Add Modal backend for audio mixdown

Co-authored-by: Igor Loskutov <igor.loskutoff@gmail.com>

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## [0.28.0](https://github.com/Monadical-SAS/reflector/compare/v0.27.0...v0.28.0) (2026-01-20)
+
+
+### Features
+
+* worker affinity ([#819](https://github.com/Monadical-SAS/reflector/issues/819)) ([3b6540e](https://github.com/Monadical-SAS/reflector/commit/3b6540eae5b597449f98661bdf15483b77be3268))
+
 ## [0.27.0](https://github.com/Monadical-SAS/reflector/compare/v0.26.0...v0.27.0) (2025-12-26)
 
 

docker-compose.yml

@@ -34,7 +34,7 @@ services:
     environment:
       ENTRYPOINT: beat
 
-  hatchet-worker:
+  hatchet-worker-cpu:
     build:
       context: server
     volumes:
@@ -43,7 +43,20 @@ services:
     env_file:
       - ./server/.env
     environment:
-      ENTRYPOINT: hatchet-worker
+      ENTRYPOINT: hatchet-worker-cpu
+    depends_on:
+      hatchet:
+        condition: service_healthy
+  hatchet-worker-llm:
+    build:
+      context: server
+    volumes:
+      - ./server/:/app/
+      - /app/.venv
+    env_file:
+      - ./server/.env
+    environment:
+      ENTRYPOINT: hatchet-worker-llm
     depends_on:
       hatchet:
         condition: service_healthy

gpu/modal_deployments/deploy-all.sh

@@ -131,6 +131,15 @@ if [ -z "$DIARIZER_URL" ]; then
 fi
 echo "  -> $DIARIZER_URL"
 
+echo ""
+echo "Deploying mixdown (CPU audio processing)..."
+MIXDOWN_URL=$(modal deploy reflector_mixdown.py 2>&1 | grep -o 'https://[^ ]*web.modal.run' | head -1)
+if [ -z "$MIXDOWN_URL" ]; then
+    echo "Error: Failed to deploy mixdown. Check Modal dashboard for details."
+    exit 1
+fi
+echo "  -> $MIXDOWN_URL"
+
 # --- Output Configuration ---
 echo ""
 echo "=========================================="
@@ -147,4 +156,8 @@ echo ""
 echo "DIARIZATION_BACKEND=modal"
 echo "DIARIZATION_URL=$DIARIZER_URL"
 echo "DIARIZATION_MODAL_API_KEY=$API_KEY"
+echo ""
+echo "MIXDOWN_BACKEND=modal"
+echo "MIXDOWN_URL=$MIXDOWN_URL"
+echo "MIXDOWN_MODAL_API_KEY=$API_KEY"
 echo "# --- End Modal Configuration ---"

gpu/modal_deployments/reflector_mixdown.py

@@ -0,0 +1,379 @@
+"""
+Reflector GPU backend - audio mixdown
+======================================
+
+CPU-intensive audio mixdown service for combining multiple audio tracks.
+Uses PyAV filter graph (amix) for high-quality audio mixing.
+"""
+
+import os
+import tempfile
+import time
+from fractions import Fraction
+
+import modal
+
+MIXDOWN_TIMEOUT = 900  # 15 minutes
+SCALEDOWN_WINDOW = 60  # 1 minute idle before shutdown
+
+app = modal.App("reflector-mixdown")
+
+# CPU-based image (no GPU needed for audio processing)
+image = (
+    modal.Image.debian_slim(python_version="3.12")
+    .apt_install("ffmpeg")  # Required by PyAV
+    .pip_install(
+        "av==13.1.0",  # PyAV for audio processing
+        "requests==2.32.3",  # HTTP for presigned URL downloads/uploads
+        "fastapi==0.115.12",  # API framework
+    )
+)
+
+
+@app.function(
+    cpu=4.0,  # 4 CPU cores for audio processing
+    timeout=MIXDOWN_TIMEOUT,
+    scaledown_window=SCALEDOWN_WINDOW,
+    secrets=[modal.Secret.from_name("reflector-gpu")],
+    image=image,
+)
+@modal.concurrent(max_inputs=10)
+@modal.asgi_app()
+def web():
+    import logging
+    import secrets
+    import shutil
+
+    import av
+    import requests
+    from av.audio.resampler import AudioResampler
+    from fastapi import Depends, FastAPI, HTTPException, status
+    from fastapi.security import OAuth2PasswordBearer
+    from pydantic import BaseModel
+
+    # Setup logging
+    logging.basicConfig(
+        level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
+    )
+    logger = logging.getLogger(__name__)
+
+    app = FastAPI()
+    oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
+
+    # Validate API key exists at startup
+    API_KEY = os.environ.get("REFLECTOR_GPU_APIKEY")
+    if not API_KEY:
+        raise RuntimeError("REFLECTOR_GPU_APIKEY not configured in Modal secrets")
+
+    def apikey_auth(apikey: str = Depends(oauth2_scheme)):
+        # Use constant-time comparison to prevent timing attacks
+        if secrets.compare_digest(apikey, API_KEY):
+            return
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail="Invalid API key",
+            headers={"WWW-Authenticate": "Bearer"},
+        )
+
+    class MixdownRequest(BaseModel):
+        track_urls: list[str]
+        output_url: str
+        target_sample_rate: int = 48000
+        expected_duration_sec: float | None = None
+
+    class MixdownResponse(BaseModel):
+        duration_ms: float
+        tracks_mixed: int
+        audio_uploaded: bool
+
+    def download_track(url: str, temp_dir: str, index: int) -> str:
+        """Download track from presigned URL to temp file using streaming."""
+        logger.info(f"Downloading track {index + 1}")
+        response = requests.get(url, stream=True, timeout=300)
+
+        if response.status_code == 404:
+            raise HTTPException(status_code=404, detail=f"Track {index} not found")
+        if response.status_code == 403:
+            raise HTTPException(
+                status_code=403, detail=f"Track {index} presigned URL expired"
+            )
+
+        response.raise_for_status()
+
+        temp_path = os.path.join(temp_dir, f"track_{index}.webm")
+        total_bytes = 0
+        with open(temp_path, "wb") as f:
+            for chunk in response.iter_content(chunk_size=8192):
+                if chunk:
+                    f.write(chunk)
+                    total_bytes += len(chunk)
+
+        logger.info(f"Track {index + 1} downloaded: {total_bytes} bytes")
+        return temp_path
+
+    def mixdown_tracks_modal(
+        track_paths: list[str],
+        output_path: str,
+        target_sample_rate: int,
+        expected_duration_sec: float | None,
+        logger,
+    ) -> float:
+        """Mix multiple audio tracks using PyAV filter graph.
+
+        Args:
+            track_paths: List of local file paths to audio tracks
+            output_path: Local path for output MP3 file
+            target_sample_rate: Sample rate for output (Hz)
+            expected_duration_sec: Optional fallback duration if container metadata unavailable
+            logger: Logger instance for progress tracking
+
+        Returns:
+            Duration in milliseconds
+        """
+        logger.info(f"Starting mixdown of {len(track_paths)} tracks")
+
+        # Build PyAV filter graph: N abuffer -> amix -> aformat -> sink
+        graph = av.filter.Graph()
+        inputs = []
+
+        for idx in range(len(track_paths)):
+            args = (
+                f"time_base=1/{target_sample_rate}:"
+                f"sample_rate={target_sample_rate}:"
+                f"sample_fmt=s32:"
+                f"channel_layout=stereo"
+            )
+            in_ctx = graph.add("abuffer", args=args, name=f"in{idx}")
+            inputs.append(in_ctx)
+
+        mixer = graph.add("amix", args=f"inputs={len(inputs)}:normalize=0", name="mix")
+        fmt = graph.add(
+            "aformat",
+            args=f"sample_fmts=s32:channel_layouts=stereo:sample_rates={target_sample_rate}",
+            name="fmt",
+        )
+        sink = graph.add("abuffersink", name="out")
+
+        # Connect inputs to mixer (no delays for Modal implementation)
+        for idx, in_ctx in enumerate(inputs):
+            in_ctx.link_to(mixer, 0, idx)
+
+        mixer.link_to(fmt)
+        fmt.link_to(sink)
+        graph.configure()
+
+        # Open all containers
+        containers = []
+        try:
+            for i, path in enumerate(track_paths):
+                try:
+                    c = av.open(path)
+                    containers.append(c)
+                except Exception as e:
+                    logger.warning(
+                        f"Failed to open container {i}: {e}",
+                    )
+
+            if not containers:
+                raise ValueError("Could not open any track containers")
+
+            # Calculate total duration for progress reporting
+            max_duration_sec = 0.0
+            for c in containers:
+                if c.duration is not None:
+                    dur_sec = c.duration / av.time_base
+                    max_duration_sec = max(max_duration_sec, dur_sec)
+            if max_duration_sec == 0.0 and expected_duration_sec:
+                max_duration_sec = expected_duration_sec
+
+            # Setup output container
+            out_container = av.open(output_path, "w", format="mp3")
+            out_stream = out_container.add_stream("libmp3lame", rate=target_sample_rate)
+
+            decoders = [c.decode(audio=0) for c in containers]
+            active = [True] * len(decoders)
+            resamplers = [
+                AudioResampler(format="s32", layout="stereo", rate=target_sample_rate)
+                for _ in decoders
+            ]
+
+            current_max_time = 0.0
+            last_log_time = time.monotonic()
+            start_time = time.monotonic()
+
+            total_duration = 0
+
+            while any(active):
+                for i, (dec, is_active) in enumerate(zip(decoders, active)):
+                    if not is_active:
+                        continue
+                    try:
+                        frame = next(dec)
+                    except StopIteration:
+                        active[i] = False
+                        inputs[i].push(None)  # Signal end of stream
+                        continue
+
+                    if frame.sample_rate != target_sample_rate:
+                        continue
+
+                    # Progress logging (every 5 seconds)
+                    if frame.time is not None:
+                        current_max_time = max(current_max_time, frame.time)
+                        now = time.monotonic()
+                        if now - last_log_time >= 5.0:
+                            elapsed = now - start_time
+                            if max_duration_sec > 0:
+                                progress_pct = min(
+                                    100.0, (current_max_time / max_duration_sec) * 100
+                                )
+                                logger.info(
+                                    f"Mixdown progress: {progress_pct:.1f}% @ {current_max_time:.1f}s (elapsed: {elapsed:.1f}s)"
+                                )
+                            else:
+                                logger.info(
+                                    f"Mixdown progress: @ {current_max_time:.1f}s (elapsed: {elapsed:.1f}s)"
+                                )
+                            last_log_time = now
+
+                    out_frames = resamplers[i].resample(frame) or []
+                    for rf in out_frames:
+                        rf.sample_rate = target_sample_rate
+                        rf.time_base = Fraction(1, target_sample_rate)
+                        inputs[i].push(rf)
+
+                    # Pull mixed frames from sink and encode
+                    while True:
+                        try:
+                            mixed = sink.pull()
+                        except Exception:
+                            break
+                        mixed.sample_rate = target_sample_rate
+                        mixed.time_base = Fraction(1, target_sample_rate)
+
+                        # Encode and mux
+                        for packet in out_stream.encode(mixed):
+                            out_container.mux(packet)
+                            total_duration += packet.duration
+
+            # Flush remaining frames from filter graph
+            while True:
+                try:
+                    mixed = sink.pull()
+                except Exception:
+                    break
+                mixed.sample_rate = target_sample_rate
+                mixed.time_base = Fraction(1, target_sample_rate)
+
+                for packet in out_stream.encode(mixed):
+                    out_container.mux(packet)
+                    total_duration += packet.duration
+
+            # Flush encoder
+            for packet in out_stream.encode():
+                out_container.mux(packet)
+                total_duration += packet.duration
+
+            # Calculate duration in milliseconds
+            if total_duration > 0:
+                # Use the same calculation as AudioFileWriterProcessor
+                duration_ms = round(
+                    float(total_duration * out_stream.time_base * 1000), 2
+                )
+            else:
+                duration_ms = 0.0
+
+            out_container.close()
+            logger.info(f"Mixdown complete: duration={duration_ms}ms")
+
+        finally:
+            # Cleanup all containers
+            for c in containers:
+                if c is not None:
+                    try:
+                        c.close()
+                    except Exception:
+                        pass
+
+        return duration_ms
+
+    @app.post("/v1/audio/mixdown", dependencies=[Depends(apikey_auth)])
+    def mixdown(request: MixdownRequest) -> MixdownResponse:
+        """Mix multiple audio tracks into a single MP3 file.
+
+        Tracks are downloaded from presigned S3 URLs, mixed using PyAV,
+        and uploaded to a presigned S3 PUT URL.
+        """
+        if not request.track_urls:
+            raise HTTPException(status_code=400, detail="No track URLs provided")
+
+        logger.info(f"Mixdown request: {len(request.track_urls)} tracks")
+
+        temp_dir = tempfile.mkdtemp()
+        temp_files = []
+        output_mp3_path = None
+
+        try:
+            # Download all tracks
+            for i, url in enumerate(request.track_urls):
+                temp_path = download_track(url, temp_dir, i)
+                temp_files.append(temp_path)
+
+            # Mix tracks
+            output_mp3_path = os.path.join(temp_dir, "mixed.mp3")
+            duration_ms = mixdown_tracks_modal(
+                temp_files,
+                output_mp3_path,
+                request.target_sample_rate,
+                request.expected_duration_sec,
+                logger,
+            )
+
+            # Upload result to S3
+            logger.info("Uploading result to S3")
+            file_size = os.path.getsize(output_mp3_path)
+            with open(output_mp3_path, "rb") as f:
+                upload_response = requests.put(
+                    request.output_url, data=f, timeout=300
+                )
+
+            if upload_response.status_code == 403:
+                raise HTTPException(
+                    status_code=403, detail="Output presigned URL expired"
+                )
+
+            upload_response.raise_for_status()
+            logger.info(f"Upload complete: {file_size} bytes")
+
+            return MixdownResponse(
+                duration_ms=duration_ms,
+                tracks_mixed=len(request.track_urls),
+                audio_uploaded=True,
+            )
+
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error(f"Mixdown failed: {e}", exc_info=True)
+            raise HTTPException(status_code=500, detail=f"Mixdown failed: {str(e)}")
+
+        finally:
+            # Cleanup temp files
+            for temp_path in temp_files:
+                try:
+                    os.unlink(temp_path)
+                except Exception as e:
+                    logger.warning(f"Failed to cleanup temp file {temp_path}: {e}")
+
+            if output_mp3_path and os.path.exists(output_mp3_path):
+                try:
+                    os.unlink(output_mp3_path)
+                except Exception as e:
+                    logger.warning(f"Failed to cleanup output file {output_mp3_path}: {e}")
+
+            try:
+                shutil.rmtree(temp_dir)
+            except Exception as e:
+                logger.warning(f"Failed to cleanup temp directory {temp_dir}: {e}")
+
+    return app

server/reflector/hatchet/run_workers.py

@@ -1,77 +0,0 @@
-"""
-Run Hatchet workers for the multitrack pipeline.
-Runs as a separate process, just like Celery workers.
-
-Usage:
-    uv run -m reflector.hatchet.run_workers
-
-    # Or via docker:
-    docker compose exec server uv run -m reflector.hatchet.run_workers
-"""
-
-import signal
-import sys
-
-from hatchet_sdk.rate_limit import RateLimitDuration
-
-from reflector.hatchet.constants import LLM_RATE_LIMIT_KEY, LLM_RATE_LIMIT_PER_SECOND
-from reflector.logger import logger
-from reflector.settings import settings
-
-
-def main() -> None:
-    """Start Hatchet worker polling."""
-    if not settings.HATCHET_ENABLED:
-        logger.error("HATCHET_ENABLED is False, not starting workers")
-        sys.exit(1)
-
-    if not settings.HATCHET_CLIENT_TOKEN:
-        logger.error("HATCHET_CLIENT_TOKEN is not set")
-        sys.exit(1)
-
-    logger.info(
-        "Starting Hatchet workers",
-        debug=settings.HATCHET_DEBUG,
-    )
-
-    # Import here (not top-level) - workflow modules call HatchetClientManager.get_client()
-    # at module level because Hatchet SDK decorators (@workflow.task) bind at import time.
-    # Can't use lazy init: decorators need the client object when function is defined.
-    from reflector.hatchet.client import HatchetClientManager  # noqa: PLC0415
-    from reflector.hatchet.workflows import (  # noqa: PLC0415
-        daily_multitrack_pipeline,
-        subject_workflow,
-        topic_chunk_workflow,
-        track_workflow,
-    )
-
-    hatchet = HatchetClientManager.get_client()
-
-    hatchet.rate_limits.put(
-        LLM_RATE_LIMIT_KEY, LLM_RATE_LIMIT_PER_SECOND, RateLimitDuration.SECOND
-    )
-
-    worker = hatchet.worker(
-        "reflector-pipeline-worker",
-        workflows=[
-            daily_multitrack_pipeline,
-            subject_workflow,
-            topic_chunk_workflow,
-            track_workflow,
-        ],
-    )
-
-    def shutdown_handler(signum: int, frame) -> None:
-        logger.info("Received shutdown signal, stopping workers...")
-        # Worker cleanup happens automatically on exit
-        sys.exit(0)
-
-    signal.signal(signal.SIGINT, shutdown_handler)
-    signal.signal(signal.SIGTERM, shutdown_handler)
-
-    logger.info("Starting Hatchet worker polling...")
-    worker.start()
-
-
-if __name__ == "__main__":
-    main()

server/reflector/hatchet/run_workers_cpu.py

@@ -0,0 +1,48 @@
+"""
+CPU-heavy worker pool for audio processing tasks.
+Handles ONLY: mixdown_tracks
+
+Configuration:
+- slots=1: Only mixdown (already serialized globally with max_runs=1)
+- Worker affinity: pool=cpu-heavy
+"""
+
+from reflector.hatchet.client import HatchetClientManager
+from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+    daily_multitrack_pipeline,
+)
+from reflector.logger import logger
+from reflector.settings import settings
+
+
+def main():
+    if not settings.HATCHET_ENABLED:
+        logger.error("HATCHET_ENABLED is False, not starting CPU workers")
+        return
+
+    hatchet = HatchetClientManager.get_client()
+
+    logger.info(
+        "Starting Hatchet CPU worker pool (mixdown only)",
+        worker_name="cpu-worker-pool",
+        slots=1,
+        labels={"pool": "cpu-heavy"},
+    )
+
+    cpu_worker = hatchet.worker(
+        "cpu-worker-pool",
+        slots=1,  # Only 1 mixdown at a time (already serialized globally)
+        labels={
+            "pool": "cpu-heavy",
+        },
+        workflows=[daily_multitrack_pipeline],
+    )
+
+    try:
+        cpu_worker.start()
+    except KeyboardInterrupt:
+        logger.info("Received shutdown signal, stopping CPU workers...")
+
+
+if __name__ == "__main__":
+    main()

server/reflector/hatchet/run_workers_llm.py

@@ -0,0 +1,56 @@
+"""
+LLM/I/O worker pool for all non-CPU tasks.
+Handles: all tasks except mixdown_tracks (transcription, LLM inference, orchestration)
+"""
+
+from reflector.hatchet.client import HatchetClientManager
+from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+    daily_multitrack_pipeline,
+)
+from reflector.hatchet.workflows.subject_processing import subject_workflow
+from reflector.hatchet.workflows.topic_chunk_processing import topic_chunk_workflow
+from reflector.hatchet.workflows.track_processing import track_workflow
+from reflector.logger import logger
+from reflector.settings import settings
+
+SLOTS = 10
+WORKER_NAME = "llm-worker-pool"
+POOL = "llm-io"
+
+
+def main():
+    if not settings.HATCHET_ENABLED:
+        logger.error("HATCHET_ENABLED is False, not starting LLM workers")
+        return
+
+    hatchet = HatchetClientManager.get_client()
+
+    logger.info(
+        "Starting Hatchet LLM worker pool (all tasks except mixdown)",
+        worker_name=WORKER_NAME,
+        slots=SLOTS,
+        labels={"pool": POOL},
+    )
+
+    llm_worker = hatchet.worker(
+        WORKER_NAME,
+        slots=SLOTS,  # not all slots are probably used
+        labels={
+            "pool": POOL,
+        },
+        workflows=[
+            daily_multitrack_pipeline,
+            topic_chunk_workflow,
+            subject_workflow,
+            track_workflow,
+        ],
+    )
+
+    try:
+        llm_worker.start()
+    except KeyboardInterrupt:
+        logger.info("Received shutdown signal, stopping LLM workers...")
+
+
+if __name__ == "__main__":
+    main()

server/reflector/hatchet/workflows/daily_multitrack_pipeline.py

@@ -23,7 +23,12 @@ from pathlib import Path
 from typing import Any, Callable, Coroutine, Protocol, TypeVar
 
 import httpx
-from hatchet_sdk import Context
+from hatchet_sdk import (
+    ConcurrencyExpression,
+    ConcurrencyLimitStrategy,
+    Context,
+)
+from hatchet_sdk.labels import DesiredWorkerLabel
 from pydantic import BaseModel
 
 from reflector.dailyco_api.client import DailyApiClient
@@ -467,10 +472,24 @@ async def process_tracks(input: PipelineInput, ctx: Context) -> ProcessTracksRes
     parents=[process_tracks],
     execution_timeout=timedelta(seconds=TIMEOUT_AUDIO),
     retries=3,
+    desired_worker_labels={
+        "pool": DesiredWorkerLabel(
+            value="cpu-heavy",
+            required=True,
+            weight=100,
+        ),
+    },
+    concurrency=[
+        ConcurrencyExpression(
+            expression="'mixdown-global'",
+            max_runs=1,  # serialize mixdown to prevent resource contention
+            limit_strategy=ConcurrencyLimitStrategy.GROUP_ROUND_ROBIN,  # Queue
+        )
+    ],
 )
 @with_error_handling(TaskName.MIXDOWN_TRACKS)
 async def mixdown_tracks(input: PipelineInput, ctx: Context) -> MixdownResult:
-    """Mix all padded tracks into single audio file using PyAV (same as Celery)."""
+    """Mix all padded tracks into single audio file using PyAV or Modal backend."""
     ctx.log("mixdown_tracks: mixing padded tracks into single audio file")
 
     track_result = ctx.task_output(process_tracks)
@@ -494,7 +513,7 @@ async def mixdown_tracks(input: PipelineInput, ctx: Context) -> MixdownResult:
 
     storage = _spawn_storage()
 
-    # Presign URLs on demand (avoids stale URLs on workflow replay)
+    # Presign URLs for padded tracks (same expiration for both backends)
     padded_urls = []
     for track_info in padded_tracks:
         if track_info.key:
@@ -515,33 +534,104 @@ async def mixdown_tracks(input: PipelineInput, ctx: Context) -> MixdownResult:
         logger.error("Mixdown failed - no decodable audio frames found")
         raise ValueError("No decodable audio frames in any track")
 
-    output_path = tempfile.mktemp(suffix=".mp3")
+    output_key = f"{input.transcript_id}/audio.mp3"
-    duration_ms_callback_capture_container = [0.0]
 
-    async def capture_duration(d):
+    # Conditional: Modal or local backend
-        duration_ms_callback_capture_container[0] = d
+    if settings.MIXDOWN_BACKEND == "modal":
+        ctx.log("mixdown_tracks: using Modal backend")
 
-    writer = AudioFileWriterProcessor(path=output_path, on_duration=capture_duration)
+        # Presign PUT URL for output (Modal will upload directly)
+        output_url = await storage.get_file_url(
+            output_key,
+            operation="put_object",
+            expires_in=PRESIGNED_URL_EXPIRATION_SECONDS,
+        )
 
-    await mixdown_tracks_pyav(
+        from reflector.processors.audio_mixdown_modal import (  # noqa: PLC0415
-        valid_urls,
+            AudioMixdownModalProcessor,
-        writer,
+        )
-        target_sample_rate,
-        offsets_seconds=None,
-        logger=logger,
-        progress_callback=make_audio_progress_logger(ctx, TaskName.MIXDOWN_TRACKS),
-        expected_duration_sec=recording_duration if recording_duration > 0 else None,
-    )
-    await writer.flush()
 
-    file_size = Path(output_path).stat().st_size
+        try:
-    storage_path = f"{input.transcript_id}/audio.mp3"
+            processor = AudioMixdownModalProcessor()
+            result = await processor.mixdown(
+                track_urls=valid_urls,
+                output_url=output_url,
+                target_sample_rate=target_sample_rate,
+                expected_duration_sec=recording_duration
+                if recording_duration > 0
+                else None,
+            )
+            duration_ms = result.duration_ms
+            tracks_mixed = result.tracks_mixed
 
-    with open(output_path, "rb") as mixed_file:
+            ctx.log(
-        await storage.put_file(storage_path, mixed_file)
+                f"mixdown_tracks: Modal returned duration={duration_ms}ms, tracks={tracks_mixed}"
+            )
+        except httpx.HTTPStatusError as e:
+            error_detail = e.response.text if hasattr(e.response, "text") else str(e)
+            logger.error(
+                "[Hatchet] Modal mixdown HTTP error",
+                transcript_id=input.transcript_id,
+                status_code=e.response.status_code if hasattr(e, "response") else None,
+                error=error_detail,
+            )
+            raise RuntimeError(
+                f"Modal mixdown failed with HTTP {e.response.status_code}: {error_detail}"
+            )
+        except httpx.TimeoutException:
+            logger.error(
+                "[Hatchet] Modal mixdown timeout",
+                transcript_id=input.transcript_id,
+                timeout=settings.MIXDOWN_TIMEOUT,
+            )
+            raise RuntimeError(
+                f"Modal mixdown timeout after {settings.MIXDOWN_TIMEOUT}s"
+            )
+        except ValueError as e:
+            logger.error(
+                "[Hatchet] Modal mixdown validation error",
+                transcript_id=input.transcript_id,
+                error=str(e),
+            )
+            raise
+    else:
+        ctx.log("mixdown_tracks: using local backend")
 
-    Path(output_path).unlink(missing_ok=True)
+        # Existing local implementation
+        output_path = tempfile.mktemp(suffix=".mp3")
+        duration_ms_callback_capture_container = [0.0]
 
+        async def capture_duration(d):
+            duration_ms_callback_capture_container[0] = d
+
+        writer = AudioFileWriterProcessor(
+            path=output_path, on_duration=capture_duration
+        )
+
+        await mixdown_tracks_pyav(
+            valid_urls,
+            writer,
+            target_sample_rate,
+            offsets_seconds=None,
+            logger=logger,
+            progress_callback=make_audio_progress_logger(ctx, TaskName.MIXDOWN_TRACKS),
+            expected_duration_sec=recording_duration
+            if recording_duration > 0
+            else None,
+        )
+        await writer.flush()
+
+        file_size = Path(output_path).stat().st_size
+        with open(output_path, "rb") as mixed_file:
+            await storage.put_file(output_key, mixed_file)
+
+        Path(output_path).unlink(missing_ok=True)
+        duration_ms = duration_ms_callback_capture_container[0]
+        tracks_mixed = len(valid_urls)
+
+        ctx.log(f"mixdown_tracks: local mixdown uploaded {file_size} bytes")
+
+    # Update DB (same for both backends)
     async with fresh_db_connection():
         from reflector.db.transcripts import transcripts_controller  # noqa: PLC0415
 
@@ -551,12 +641,12 @@ async def mixdown_tracks(input: PipelineInput, ctx: Context) -> MixdownResult:
                 transcript, {"audio_location": "storage"}
             )
 
-    ctx.log(f"mixdown_tracks complete: uploaded {file_size} bytes to {storage_path}")
+    ctx.log(f"mixdown_tracks complete: uploaded to {output_key}")
 
     return MixdownResult(
-        audio_key=storage_path,
+        audio_key=output_key,
-        duration=duration_ms_callback_capture_container[0],
+        duration=duration_ms,
-        tracks_mixed=len(valid_urls),
+        tracks_mixed=tracks_mixed,
     )
 
 

server/reflector/hatchet/workflows/topic_chunk_processing.py

@@ -7,7 +7,11 @@ Spawned dynamically by detect_topics via aio_run_many() for parallel processing.
 
 from datetime import timedelta
 
-from hatchet_sdk import ConcurrencyExpression, ConcurrencyLimitStrategy, Context
+from hatchet_sdk import (
+    ConcurrencyExpression,
+    ConcurrencyLimitStrategy,
+    Context,
+)
 from hatchet_sdk.rate_limit import RateLimit
 from pydantic import BaseModel
 
@@ -34,11 +38,13 @@ hatchet = HatchetClientManager.get_client()
 topic_chunk_workflow = hatchet.workflow(
     name="TopicChunkProcessing",
     input_validator=TopicChunkInput,
-    concurrency=ConcurrencyExpression(
+    concurrency=[
-        expression="'global'",  # constant string = global limit across all runs
+        ConcurrencyExpression(
-        max_runs=20,
+            expression="'global'",  # constant string = global limit across all runs
-        limit_strategy=ConcurrencyLimitStrategy.GROUP_ROUND_ROBIN,
+            max_runs=20,
-    ),
+            limit_strategy=ConcurrencyLimitStrategy.GROUP_ROUND_ROBIN,
+        )
+    ],
 )
 
 

server/reflector/processors/audio_mixdown_modal.py

@@ -0,0 +1,89 @@
+"""
+Modal.com backend for audio mixdown.
+
+Uses Modal's CPU containers to offload audio mixing from Hatchet workers.
+Communicates via presigned S3 URLs for both input and output.
+"""
+
+import httpx
+from pydantic import BaseModel
+
+from reflector.settings import settings
+
+
+class MixdownResponse(BaseModel):
+    """Response from Modal mixdown endpoint."""
+
+    duration_ms: float
+    tracks_mixed: int
+    audio_uploaded: bool
+
+
+class AudioMixdownModalProcessor:
+    """Audio mixdown processor using Modal.com CPU backend.
+
+    Sends track URLs (presigned GET) and output URL (presigned PUT) to Modal.
+    Modal handles download, mixdown via PyAV, and upload.
+    """
+
+    def __init__(self, modal_api_key: str | None = None):
+        if not settings.MIXDOWN_URL:
+            raise ValueError("MIXDOWN_URL required to use AudioMixdownModalProcessor")
+
+        self.mixdown_url = settings.MIXDOWN_URL + "/v1"
+        self.timeout = settings.MIXDOWN_TIMEOUT
+        self.modal_api_key = modal_api_key or settings.MIXDOWN_MODAL_API_KEY
+
+        if not self.modal_api_key:
+            raise ValueError(
+                "MIXDOWN_MODAL_API_KEY required to use AudioMixdownModalProcessor"
+            )
+
+    async def mixdown(
+        self,
+        track_urls: list[str],
+        output_url: str,
+        target_sample_rate: int,
+        expected_duration_sec: float | None = None,
+    ) -> MixdownResponse:
+        """Mix multiple audio tracks via Modal backend.
+
+        Args:
+            track_urls: List of presigned GET URLs for audio tracks (non-empty)
+            output_url: Presigned PUT URL for output MP3
+            target_sample_rate: Sample rate for output (Hz, must be positive)
+            expected_duration_sec: Optional fallback duration if container metadata unavailable
+
+        Returns:
+            MixdownResponse with duration_ms, tracks_mixed, audio_uploaded
+
+        Raises:
+            ValueError: If track_urls is empty or target_sample_rate invalid
+            httpx.HTTPStatusError: On HTTP errors (404, 403, 500, etc.)
+            httpx.TimeoutException: On timeout
+        """
+        # Validate inputs
+        if not track_urls:
+            raise ValueError("track_urls cannot be empty")
+        if target_sample_rate <= 0:
+            raise ValueError(
+                f"target_sample_rate must be positive, got {target_sample_rate}"
+            )
+        if expected_duration_sec is not None and expected_duration_sec < 0:
+            raise ValueError(
+                f"expected_duration_sec cannot be negative, got {expected_duration_sec}"
+            )
+
+        async with httpx.AsyncClient(timeout=self.timeout) as client:
+            response = await client.post(
+                f"{self.mixdown_url}/audio/mixdown",
+                headers={"Authorization": f"Bearer {self.modal_api_key}"},
+                json={
+                    "track_urls": track_urls,
+                    "output_url": output_url,
+                    "target_sample_rate": target_sample_rate,
+                    "expected_duration_sec": expected_duration_sec,
+                },
+            )
+            response.raise_for_status()
+            return MixdownResponse(**response.json())

server/reflector/settings.py

@@ -98,6 +98,17 @@ class Settings(BaseSettings):
     # Diarization: local pyannote.audio
     DIARIZATION_PYANNOTE_AUTH_TOKEN: str | None = None
 
+    # Audio Mixdown
+    # backends:
+    #   - local: in-process PyAV mixdown (runs in same process as Hatchet worker)
+    #   - modal: HTTP API client to Modal.com CPU container
+    MIXDOWN_BACKEND: str = "local"
+    MIXDOWN_URL: str | None = None
+    MIXDOWN_TIMEOUT: int = 900  # 15 minutes
+
+    # Mixdown: modal backend
+    MIXDOWN_MODAL_API_KEY: str | None = None
+
     # Sentry
     SENTRY_DSN: str | None = None
 

server/runserver.sh

@@ -7,8 +7,10 @@ elif [ "${ENTRYPOINT}" = "worker" ]; then
     uv run celery -A reflector.worker.app worker --loglevel=info
 elif [ "${ENTRYPOINT}" = "beat" ]; then
     uv run celery -A reflector.worker.app beat --loglevel=info
-elif [ "${ENTRYPOINT}" = "hatchet-worker" ]; then
+elif [ "${ENTRYPOINT}" = "hatchet-worker-cpu" ]; then
-    uv run python -m reflector.hatchet.run_workers
+    uv run python -m reflector.hatchet.run_workers_cpu
+elif [ "${ENTRYPOINT}" = "hatchet-worker-llm" ]; then
+    uv run python -m reflector.hatchet.run_workers_llm
 else
     echo "Unknown command"
 fi

www/app/(app)/transcripts/[transcriptId]/_components/TopicList.tsx

@@ -1,12 +1,12 @@
 import React, { useState, useEffect } from "react";
 import ScrollToBottom from "../../scrollToBottom";
 import { Topic } from "../../webSocketTypes";
-import { Box, Flex, Text } from "@chakra-ui/react";
+import useParticipants from "../../useParticipants";
-import { formatTime } from "../../../../lib/time";
+import { Box, Flex, Text, Accordion } from "@chakra-ui/react";
-import { getTopicColor } from "../../../../lib/topicColors";
+import { TopicItem } from "./TopicItem";
 import { TranscriptStatus } from "../../../../lib/transcript";
+
 import { featureEnabled } from "../../../../lib/features";
-import { TOPICS_SCROLL_DIV_ID } from "./constants";
 
 type TopicListProps = {
   topics: Topic[];
@@ -18,7 +18,6 @@ type TopicListProps = {
   transcriptId: string;
   status: TranscriptStatus | null;
   currentTranscriptText: any;
-  onTopicClick?: (topicId: string) => void;
 };
 
 export function TopicList({
@@ -28,13 +27,30 @@ export function TopicList({
   transcriptId,
   status,
   currentTranscriptText,
-  onTopicClick,
 }: TopicListProps) {
   const [activeTopic, setActiveTopic] = useActiveTopic;
-  const [hoveredTopicId, setHoveredTopicId] = useState<string | null>(null);
   const [autoscrollEnabled, setAutoscrollEnabled] = useState<boolean>(true);
+  const participants = useParticipants(transcriptId);
 
-  const toggleScroll = (element: HTMLElement) => {
+  const scrollToTopic = () => {
+    const topicDiv = document.getElementById(`topic-${activeTopic?.id}`);
+
+    setTimeout(() => {
+      topicDiv?.scrollIntoView({
+        behavior: "smooth",
+        block: "start",
+        inline: "nearest",
+      });
+    }, 200);
+  };
+
+  useEffect(() => {
+    if (activeTopic && autoscroll) scrollToTopic();
+  }, [activeTopic, autoscroll]);
+
+  // scroll top is not rounded, heights are, so exact match won't work.
+  // https://developer.mozilla.org/en-US/docs/Web/API/Element/scrollHeight#determine_if_an_element_has_been_totally_scrolled
+  const toggleScroll = (element) => {
     const bottom =
       Math.abs(
         element.scrollHeight - element.clientHeight - element.scrollTop,
@@ -45,19 +61,14 @@ export function TopicList({
       setAutoscrollEnabled(true);
     }
   };
-
+  const handleScroll = (e) => {
-  const handleScroll = (e: React.UIEvent<HTMLDivElement>) => {
+    toggleScroll(e.target);
-    toggleScroll(e.target as HTMLElement);
-  };
-
-  const scrollToBottom = () => {
-    const topicsDiv = document.getElementById(TOPICS_SCROLL_DIV_ID);
-    if (topicsDiv) topicsDiv.scrollTop = topicsDiv.scrollHeight;
   };
 
   useEffect(() => {
     if (autoscroll) {
-      const topicsDiv = document.getElementById(TOPICS_SCROLL_DIV_ID);
+      const topicsDiv = document.getElementById("scroll-div");
+
       topicsDiv && toggleScroll(topicsDiv);
     }
   }, [activeTopic, autoscroll]);
@@ -66,41 +77,37 @@ export function TopicList({
     if (autoscroll && autoscrollEnabled) scrollToBottom();
   }, [topics.length, currentTranscriptText]);
 
+  const scrollToBottom = () => {
+    const topicsDiv = document.getElementById("scroll-div");
+
+    if (topicsDiv) topicsDiv.scrollTop = topicsDiv.scrollHeight;
+  };
+
+  const getSpeakerName = (speakerNumber: number) => {
+    if (!participants.response) return;
+    return (
+      participants.response.find(
+        (participant) => participant.speaker == speakerNumber,
+      )?.name || `Speaker ${speakerNumber}`
+    );
+  };
+
+  const requireLogin = featureEnabled("requireLogin");
+
   useEffect(() => {
     if (autoscroll) {
       setActiveTopic(topics[topics.length - 1]);
     }
   }, [topics, autoscroll]);
 
-  const handleTopicClick = (topic: Topic) => {
-    setActiveTopic(topic);
-    if (onTopicClick) {
-      onTopicClick(topic.id);
-    }
-  };
-
-  const handleTopicMouseEnter = (topic: Topic) => {
-    setHoveredTopicId(topic.id);
-    // If already active, toggle off when mousing over
-    if (activeTopic?.id === topic.id) {
-      setActiveTopic(null);
-    } else {
-      setActiveTopic(topic);
-    }
-  };
-
-  const handleTopicMouseLeave = () => {
-    setHoveredTopicId(null);
-  };
-
-  const requireLogin = featureEnabled("requireLogin");
-
   return (
     <Flex
-      position="relative"
+      position={"relative"}
-      w="full"
+      w={"100%"}
-      h="200px"
+      h={"95%"}
-      flexDirection="column"
+      flexDirection={"column"}
+      justify={"center"}
+      align={"center"}
       flexShrink={0}
     >
       {autoscroll && (
@@ -111,71 +118,45 @@ export function TopicList({
       )}
 
       <Box
-        id={TOPICS_SCROLL_DIV_ID}
+        id="scroll-div"
-        overflowY="auto"
+        overflowY={"auto"}
-        h="full"
+        h={"100%"}
         onScroll={handleScroll}
         width="full"
       >
         {topics.length > 0 && (
-          <Flex direction="column" gap={1} p={2}>
+          <Accordion.Root
-            {topics.map((topic, index) => {
+            multiple={false}
-              const color = getTopicColor(index);
+            collapsible={true}
-              const isActive = activeTopic?.id === topic.id;
+            value={activeTopic ? [activeTopic.id] : []}
-              const isHovered = hoveredTopicId === topic.id;
+            onValueChange={(details) => {
-
+              const selectedTopicId = details.value[0];
-              return (
+              const selectedTopic = selectedTopicId
-                <Flex
+                ? topics.find((t) => t.id === selectedTopicId)
-                  key={topic.id}
+                : null;
-                  id={`topic-${topic.id}`}
+              setActiveTopic(selectedTopic || null);
-                  gap={2}
+            }}
-                  align="center"
+          >
-                  py={1}
+            {topics.map((topic) => (
-                  px={2}
+              <TopicItem
-                  cursor="pointer"
+                key={topic.id}
-                  bg={isActive || isHovered ? "gray.100" : "transparent"}
+                topic={topic}
-                  _hover={{ bg: "gray.50" }}
+                isActive={activeTopic?.id === topic.id}
-                  onClick={() => handleTopicClick(topic)}
+                getSpeakerName={getSpeakerName}
-                  onMouseEnter={() => handleTopicMouseEnter(topic)}
+              />
-                  onMouseLeave={handleTopicMouseLeave}
+            ))}
-                >
+          </Accordion.Root>
-                  {/* Color indicator */}
-                  <Box
-                    w="12px"
-                    h="12px"
-                    borderRadius="full"
-                    bg={color}
-                    flexShrink={0}
-                  />
-
-                  {/* Topic title */}
-                  <Text
-                    flex={1}
-                    fontSize="sm"
-                    fontWeight={isActive ? "semibold" : "normal"}
-                  >
-                    {topic.title}
-                  </Text>
-
-                  {/* Timestamp */}
-                  <Text as="span" color="gray.500" fontSize="xs" flexShrink={0}>
-                    {formatTime(topic.timestamp)}
-                  </Text>
-                </Flex>
-              );
-            })}
-          </Flex>
         )}
 
         {status == "recording" && (
-          <Box textAlign="center">
+          <Box textAlign={"center"}>
             <Text>{currentTranscriptText}</Text>
           </Box>
         )}
         {(status == "recording" || status == "idle") &&
           currentTranscriptText.length == 0 &&
           topics.length == 0 && (
-            <Box textAlign="center" color="gray">
+            <Box textAlign={"center"} color="gray">
               <Text>
                 Full discussion transcript will appear here after you start
                 recording.
@@ -186,7 +167,7 @@ export function TopicList({
             </Box>
           )}
         {status == "processing" && (
-          <Box textAlign="center" color="gray">
+          <Box textAlign={"center"} color="gray">
             <Text>We are processing the recording, please wait.</Text>
             {!requireLogin && (
               <span>
@@ -196,12 +177,12 @@ export function TopicList({
           </Box>
         )}
         {status == "ended" && topics.length == 0 && (
-          <Box textAlign="center" color="gray">
+          <Box textAlign={"center"} color="gray">
             <Text>Recording has ended without topics being found.</Text>
           </Box>
         )}
         {status == "error" && (
-          <Box textAlign="center" color="gray">
+          <Box textAlign={"center"} color="gray">
             <Text>There was an error processing your recording</Text>
           </Box>
         )}

www/app/(app)/transcripts/[transcriptId]/_components/TranscriptWithGutter.tsx

@@ -1,106 +0,0 @@
-import { Box, Text, IconButton } from "@chakra-ui/react";
-import { ChevronUp } from "lucide-react";
-import { Topic } from "../../webSocketTypes";
-import { getTopicColor } from "../../../../lib/topicColors";
-import { TOPICS_SCROLL_DIV_ID } from "./constants";
-
-interface TranscriptWithGutterProps {
-  topics: Topic[];
-  getSpeakerName: (speakerNumber: number) => string | undefined;
-  onGutterClick: (topicId: string) => void;
-}
-
-export function TranscriptWithGutter({
-  topics,
-  getSpeakerName,
-  onGutterClick,
-}: TranscriptWithGutterProps) {
-  const scrollToTopics = () => {
-    // Scroll to the topic list at the top
-    const topicList = document.getElementById(TOPICS_SCROLL_DIV_ID);
-    if (topicList) {
-      topicList.scrollIntoView({
-        behavior: "smooth",
-        block: "start",
-      });
-    }
-  };
-
-  return (
-    <Box>
-      {topics.map((topic, topicIndex) => {
-        const color = getTopicColor(topicIndex);
-
-        return (
-          <Box key={topic.id} position="relative">
-            {/* Topic Header with Up Button */}
-            <Box
-              py={3}
-              px={4}
-              fontWeight="semibold"
-              fontSize="lg"
-              display="flex"
-              alignItems="center"
-              justifyContent="space-between"
-            >
-              <Text>{topic.title}</Text>
-              <IconButton
-                aria-label="Scroll to topics"
-                size="sm"
-                variant="ghost"
-                onClick={scrollToTopics}
-              >
-                <ChevronUp size={16} />
-              </IconButton>
-            </Box>
-
-            {/* Segments container with single gutter */}
-            <Box position="relative">
-              {/* Single continuous gutter for entire topic */}
-              <Box
-                className="topic-gutter"
-                position="absolute"
-                left={0}
-                top={0}
-                bottom={0}
-                width="4px"
-                bg={color}
-                cursor="pointer"
-                transition="all 0.2s"
-                _hover={{
-                  filter: "brightness(1.2)",
-                  width: "6px",
-                }}
-                onClick={() => onGutterClick(topic.id)}
-              />
-
-              {/* Segments */}
-              {topic.segments?.map((segment, segmentIndex) => (
-                <Box
-                  key={segmentIndex}
-                  id={`segment-${topic.id}-${segmentIndex}`}
-                  py={2}
-                  px={4}
-                  pl={12}
-                  _hover={{
-                    bg: "gray.50",
-                  }}
-                >
-                  {/* Segment Content */}
-                  <Text fontSize="sm">
-                    <Text as="span" fontWeight="semibold" color="gray.700">
-                      {getSpeakerName(segment.speaker) ||
-                        `Speaker ${segment.speaker}`}
-                      :
-                    </Text>{" "}
-                    {segment.text}
-                  </Text>
-                </Box>
-              ))}
-            </Box>
-          </Box>
-        );
-      })}
-    </Box>
-  );
-}

www/app/(app)/transcripts/[transcriptId]/_components/constants.ts

@@ -1 +0,0 @@
-export const TOPICS_SCROLL_DIV_ID = "topics-scroll-div";

www/app/(app)/transcripts/[transcriptId]/page.tsx

@@ -3,9 +3,7 @@ import Modal from "../modal";
 import useTopics from "../useTopics";
 import useWaveform from "../useWaveform";
 import useMp3 from "../useMp3";
-import useParticipants from "../useParticipants";
 import { TopicList } from "./_components/TopicList";
-import { TranscriptWithGutter } from "./_components/TranscriptWithGutter";
 import { Topic } from "../webSocketTypes";
 import React, { useEffect, useState, use } from "react";
 import FinalSummary from "./finalSummary";
@@ -47,91 +45,14 @@ export default function TranscriptDetails(details: TranscriptDetails) {
 
   const mp3 = useMp3(transcriptId, waiting);
   const topics = useTopics(transcriptId);
-  const participants = useParticipants(transcriptId);
   const waveform = useWaveform(
     transcriptId,
     waiting || mp3.audioDeleted === true,
   );
   const useActiveTopic = useState<Topic | null>(null);
-  const [activeTopic, setActiveTopic] = useActiveTopic;
   const [finalSummaryElement, setFinalSummaryElement] =
     useState<HTMLDivElement | null>(null);
 
-  // IntersectionObserver for active topic detection based on scroll position
-  useEffect(() => {
-    if (!topics.topics || topics.topics.length === 0) return;
-
-    const observer = new IntersectionObserver(
-      (entries) => {
-        // Find the most visible segment
-        let mostVisibleEntry: IntersectionObserverEntry | null = null;
-        let maxRatio = 0;
-
-        entries.forEach((entry) => {
-          if (entry.isIntersecting && entry.intersectionRatio > maxRatio) {
-            maxRatio = entry.intersectionRatio;
-            mostVisibleEntry = entry;
-          }
-        });
-
-        if (mostVisibleEntry) {
-          // Extract topicId from segment id (format: "segment-{topicId}-{idx}")
-          const segmentId = mostVisibleEntry.target.id;
-          const match = segmentId.match(/^segment-([^-]+)-/);
-          if (match) {
-            const topicId = match[1];
-            const topic = topics.topics?.find((t) => t.id === topicId);
-            if (topic && activeTopic?.id !== topic.id) {
-              setActiveTopic(topic);
-            }
-          }
-        }
-      },
-      {
-        threshold: [0, 0.25, 0.5, 0.75, 1],
-        rootMargin: "-20% 0px -20% 0px",
-      },
-    );
-
-    // Observe all segment elements
-    const segments = document.querySelectorAll('[id^="segment-"]');
-    segments.forEach((segment) => observer.observe(segment));
-
-    return () => observer.disconnect();
-  }, [topics.topics, activeTopic?.id, setActiveTopic]);
-
-  // Scroll handlers for bidirectional navigation
-  const handleTopicClick = (topicId: string) => {
-    // Scroll to first segment of this topic in transcript
-    const firstSegment = document.querySelector(`[id^="segment-${topicId}-"]`);
-    if (firstSegment) {
-      firstSegment.scrollIntoView({
-        behavior: "smooth",
-        block: "center",
-      });
-    }
-  };
-
-  const handleGutterClick = (topicId: string) => {
-    // Scroll to topic in list
-    const topicChip = document.getElementById(`topic-${topicId}`);
-    if (topicChip) {
-      topicChip.scrollIntoView({
-        behavior: "smooth",
-        block: "center",
-      });
-    }
-  };
-
-  const getSpeakerName = (speakerNumber: number) => {
-    if (!participants.response) return `Speaker ${speakerNumber}`;
-    return (
-      participants.response.find(
-        (participant) => participant.speaker == speakerNumber,
-      )?.name || `Speaker ${speakerNumber}`
-    );
-  };
-
   useEffect(() => {
     if (!waiting || !transcript.data) return;
 
@@ -200,7 +121,7 @@ export default function TranscriptDetails(details: TranscriptDetails) {
     <>
       <Grid
         templateColumns="1fr"
-        templateRows="auto auto"
+        templateRows="auto minmax(0, 1fr)"
         gap={4}
         mt={4}
         mb={4}
@@ -232,18 +153,18 @@ export default function TranscriptDetails(details: TranscriptDetails) {
         <Grid
           templateColumns={{ base: "minmax(0, 1fr)", md: "repeat(2, 1fr)" }}
           templateRows={{
-            base: "auto auto auto",
+            base: "auto minmax(0, 1fr) minmax(0, 1fr)",
-            md: "auto auto",
+            md: "auto minmax(0, 1fr)",
           }}
           gap={4}
           gridRowGap={2}
           padding={4}
+          paddingBottom={0}
           background="gray.bg"
           border={"2px solid"}
           borderColor={"gray.bg"}
           borderRadius={8}
         >
-          {/* Title */}
           <GridItem colSpan={{ base: 1, md: 2 }}>
             <Flex direction="column" gap={0}>
               <Flex alignItems="center" gap={2}>
@@ -266,64 +187,28 @@ export default function TranscriptDetails(details: TranscriptDetails) {
               )}
             </Flex>
           </GridItem>
-
+          <TopicList
-          {/* Left column: Topics List */}
+            topics={topics.topics || []}
-          <GridItem display="flex" flexDirection="column" gap={4} h="100%">
+            useActiveTopic={useActiveTopic}
-            <TopicList
+            autoscroll={false}
-              topics={topics.topics || []}
+            transcriptId={transcriptId}
-              useActiveTopic={useActiveTopic}
+            status={transcript.data?.status || null}
-              autoscroll={false}
+            currentTranscriptText=""
-              transcriptId={transcriptId}
+          />
-              status={transcript.data?.status || null}
-              currentTranscriptText=""
-              onTopicClick={handleTopicClick}
-            />
-
-            {/* Transcript with colored gutter (scrollable) */}
-            {topics.topics && topics.topics.length > 0 && (
-              <Box
-                overflowY="auto"
-                flex={1}
-                minH="0"
-                pr={2}
-                css={{
-                  "&::-webkit-scrollbar": {
-                    width: "8px",
-                  },
-                  "&::-webkit-scrollbar-track": {
-                    background: "transparent",
-                  },
-                  "&::-webkit-scrollbar-thumb": {
-                    background: "#CBD5E0",
-                    borderRadius: "4px",
-                  },
-                  "&::-webkit-scrollbar-thumb:hover": {
-                    background: "#A0AEC0",
-                  },
-                }}
-              >
-                <TranscriptWithGutter
-                  topics={topics.topics}
-                  getSpeakerName={getSpeakerName}
-                  onGutterClick={handleGutterClick}
-                />
-              </Box>
-            )}
-          </GridItem>
-
-          {/* Right column: Final Summary */}
           {transcript.data && topics.topics ? (
-            <FinalSummary
+            <>
-              transcript={transcript.data}
+              <FinalSummary
-              topics={topics.topics}
+                transcript={transcript.data}
-              onUpdate={() => {
+                topics={topics.topics}
-                transcript.refetch().then(() => {});
+                onUpdate={() => {
-              }}
+                  transcript.refetch().then(() => {});
-              finalSummaryRef={setFinalSummaryElement}
+                }}
-            />
+                finalSummaryRef={setFinalSummaryElement}
+              />
+            </>
           ) : (
-            <Flex justify="center" alignItems="center" h="100%">
+            <Flex justify={"center"} alignItems={"center"} h={"100%"}>
-              <Flex direction="column" h="full" justify="center" align="center">
+              <div className="flex flex-col h-full justify-center content-center">
                 {transcript?.data?.status == "processing" ? (
                   <Text>Loading Transcript</Text>
                 ) : (
@@ -332,7 +217,7 @@ export default function TranscriptDetails(details: TranscriptDetails) {
                     back later
                   </Text>
                 )}
-              </Flex>
+              </div>
             </Flex>
           )}
         </Grid>

www/app/lib/topicColors.ts

@@ -1,18 +0,0 @@
-// Predefined color palette for topics
-// Colors chosen for good contrast and visual distinction
-export const TOPIC_COLORS = [
-  "#3B82F6", // blue
-  "#10B981", // green
-  "#F59E0B", // amber
-  "#EF4444", // red
-  "#8B5CF6", // violet
-  "#EC4899", // pink
-  "#14B8A6", // teal
-  "#F97316", // orange
-  "#6366F1", // indigo
-  "#84CC16", // lime
-] as const;
-
-export function getTopicColor(topicIndex: number): string {
-  return TOPIC_COLORS[topicIndex % TOPIC_COLORS.length];
-}