selfhostyourtech/reflector
2
0
Fork 0
mirror of https://github.com/Monadical-SAS/reflector.git synced 2026-02-05 02:16:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: parallelize hatchet (#804)

Browse Source 
* parallelize hatchet (no-mistakes)

* dry (no-mistakes) (minimal)

* comments

* self-review

* self-review

* self-review

* self-review

* pr comments

* pr comments

---------

Co-authored-by: Igor Loskutov <igor.loskutoff@gmail.com>
This commit is contained in:
Igor Monadical
2025-12-23 11:03:36 -05:00
committed by GitHub
parent 7c2d0698ed
commit 594bcc09e0
15 changed files with 849 additions and 287 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
server/reflector/processors/prompts.py Normal file
View File

@@ -0,0 +1,30 @@
"""
LLM prompts for transcript processing.
Extracted to a separate module to avoid circular imports when importing
from processor modules (which import LLM/settings at module level).
"""
from textwrap import dedent
TOPIC_PROMPT = dedent(
"""
Analyze the following transcript segment and extract the main topic being discussed.
Focus on the substantive content and ignore small talk or administrative chatter.
Create a title that:
- Captures the specific subject matter being discussed
- Is descriptive and self-explanatory
- Uses professional language
- Is specific rather than generic
For the summary:
- Summarize the key points in maximum two sentences
- Focus on what was discussed, decided, or accomplished
- Be concise but informative
<transcript>
{text}
</transcript>
"""
).strip()

91
server/reflector/processors/summary/prompts.py Normal file
View File

@@ -0,0 +1,91 @@
"""
LLM prompts for summary generation.
Extracted to a separate module to avoid circular imports when importing
from summary_builder.py (which imports LLM/settings at module level).
"""
from textwrap import dedent
def build_participant_instructions(participant_names: list[str]) -> str:
"""Build participant context instructions for LLM prompts."""
if not participant_names:
return ""
participants_list = ", ".join(participant_names)
return dedent(
f"""
# IMPORTANT: Participant Names
The following participants are identified in this conversation: {participants_list}
You MUST use these specific participant names when referring to people in your response.
Do NOT use generic terms like "a participant", "someone", "attendee", "Speaker 1", "Speaker 2", etc.
Always refer to people by their actual names (e.g., "John suggested..." not "A participant suggested...").
"""
).strip()
DETAILED_SUBJECT_PROMPT_TEMPLATE = dedent(
"""
Get me information about the topic "{subject}"
# RESPONSE GUIDELINES
Follow this structured approach to create the topic summary:
- Highlight important arguments, insights, or data presented.
- Outline decisions made.
- Indicate any decisions reached, including any rationale or key factors
that influenced these decisions.
- Detail action items and responsibilities.
- For each decision or unresolved issue, list specific action items agreed
upon, along with assigned individuals or teams responsible for each task.
- Specify deadlines or timelines if mentioned. For each action item,
include any deadlines or timeframes discussed for completion or follow-up.
- Mention unresolved issues or topics needing further discussion, aiding in
planning future meetings or follow-up actions.
- Do not include topic unrelated to {subject}.
# OUTPUT
Your summary should be clear, concise, and structured, covering all major
points, decisions, and action items from the meeting. It should be easy to
understand for someone not present, providing a comprehensive understanding
of what transpired and what needs to be done next. The summary should not
exceed one page to ensure brevity and focus.
"""
).strip()
PARAGRAPH_SUMMARY_PROMPT = dedent(
"""
Summarize the mentioned topic in 1 paragraph.
It will be integrated into the final summary, so just for this topic.
"""
).strip()
RECAP_PROMPT = dedent(
"""
Provide a high-level quick recap of the following meeting, fitting in one paragraph.
Do not include decisions, action items or unresolved issue, just highlight the high moments.
Just dive into the meeting, be concise and do not include unnecessary details.
As we already know it is a meeting, do not start with 'During the meeting' or equivalent.
"""
).strip()
def build_summary_markdown(recap: str, summaries: list[dict[str, str]]) -> str:
"""Build markdown summary from recap and subject summaries."""
lines: list[str] = []
if recap:
lines.append("# Quick recap")
lines.append("")
lines.append(recap)
lines.append("")
if summaries:
lines.append("# Summary")
lines.append("")
for summary in summaries:
lines.append(f"**{summary['subject']}**")
lines.append(summary["summary"])
lines.append("")
return "\n".join(lines)

95
server/reflector/processors/summary/summary_builder.py
View File

@@ -15,6 +15,13 @@ import structlog
from pydantic import BaseModel, Field
from reflector.llm import LLM
from reflector.processors.summary.prompts import (
DETAILED_SUBJECT_PROMPT_TEMPLATE,
PARAGRAPH_SUMMARY_PROMPT,
RECAP_PROMPT,
build_participant_instructions,
build_summary_markdown,
)
from reflector.settings import settings
T = TypeVar("T", bound=BaseModel)
@@ -52,50 +59,6 @@ SUBJECTS_PROMPT = dedent(
"""
).strip()
DETAILED_SUBJECT_PROMPT_TEMPLATE = dedent(
"""
Get me information about the topic "{subject}"
# RESPONSE GUIDELINES
Follow this structured approach to create the topic summary:
- Highlight important arguments, insights, or data presented.
- Outline decisions made.
- Indicate any decisions reached, including any rationale or key factors
that influenced these decisions.
- Detail action items and responsibilities.
- For each decision or unresolved issue, list specific action items agreed
upon, along with assigned individuals or teams responsible for each task.
- Specify deadlines or timelines if mentioned. For each action item,
include any deadlines or timeframes discussed for completion or follow-up.
- Mention unresolved issues or topics needing further discussion, aiding in
planning future meetings or follow-up actions.
- Do not include topic unrelated to {subject}.
# OUTPUT
Your summary should be clear, concise, and structured, covering all major
points, decisions, and action items from the meeting. It should be easy to
understand for someone not present, providing a comprehensive understanding
of what transpired and what needs to be done next. The summary should not
exceed one page to ensure brevity and focus.
"""
).strip()
PARAGRAPH_SUMMARY_PROMPT = dedent(
"""
Summarize the mentioned topic in 1 paragraph.
It will be integrated into the final summary, so just for this topic.
"""
).strip()
RECAP_PROMPT = dedent(
"""
Provide a high-level quick recap of the following meeting, fitting in one paragraph.
Do not include decisions, action items or unresolved issue, just highlight the high moments.
Just dive into the meeting, be concise and do not include unnecessary details.
As we already know it is a meeting, do not start with 'During the meeting' or equivalent.
"""
).strip()
ACTION_ITEMS_PROMPT = dedent(
"""
Identify action items from this meeting transcript. Your goal is to identify what was decided and what needs to happen next.
@@ -331,17 +294,7 @@ class SummaryBuilder:
participants_md = self.format_list_md(participants)
self.transcript += f"\n\n# Participants\n\n{participants_md}"
participants_list = ", ".join(participants)
self.participant_instructions = dedent(
f"""
# IMPORTANT: Participant Names
The following participants are identified in this conversation: {participants_list}
You MUST use these specific participant names when referring to people in your response.
Do NOT use generic terms like "a participant", "someone", "attendee", "Speaker 1", "Speaker 2", etc.
Always refer to people by their actual names (e.g., "John suggested..." not "A participant suggested...").
"""
).strip()
self.participant_instructions = build_participant_instructions(participants)
async def identify_participants(self) -> None:
"""
@@ -377,18 +330,9 @@ class SummaryBuilder:
participants_md = self.format_list_md(unique_participants)
self.transcript += f"\n\n# Participants\n\n{participants_md}"
# Set instructions that will be automatically added to all prompts
participants_list = ", ".join(unique_participants)
self.participant_instructions = dedent(
f"""
# IMPORTANT: Participant Names
The following participants are identified in this conversation: {participants_list}
You MUST use these specific participant names when referring to people in your response.
Do NOT use generic terms like "a participant", "someone", "attendee", "Speaker 1", "Speaker 2", etc.
Always refer to people by their actual names (e.g., "John suggested..." not "A participant suggested...").
"""
).strip()
self.participant_instructions = build_participant_instructions(
unique_participants
)
else:
self.logger.warning("No participants identified in the transcript")
@@ -613,22 +557,7 @@ class SummaryBuilder:
# ----------------------------------------------------------------------------
def as_markdown(self) -> str:
lines: list[str] = []
if self.recap:
lines.append("# Quick recap")
lines.append("")
lines.append(self.recap)
lines.append("")
if self.summaries:
lines.append("# Summary")
lines.append("")
for summary in self.summaries:
lines.append(f"**{summary['subject']}**")
lines.append(summary["summary"])
lines.append("")
return "\n".join(lines)
return build_summary_markdown(self.recap, self.summaries)
def format_list_md(self, data: list[str]) -> str:
return "\n".join([f"- {item}" for item in data])

25
server/reflector/processors/transcript_topic_detector.py
View File

@@ -1,35 +1,12 @@
from textwrap import dedent
from pydantic import AliasChoices, BaseModel, Field
from reflector.llm import LLM
from reflector.processors.base import Processor
from reflector.processors.prompts import TOPIC_PROMPT
from reflector.processors.types import TitleSummary, Transcript
from reflector.settings import settings
from reflector.utils.text import clean_title
TOPIC_PROMPT = dedent(
"""
Analyze the following transcript segment and extract the main topic being discussed.
Focus on the substantive content and ignore small talk or administrative chatter.
Create a title that:
- Captures the specific subject matter being discussed
- Is descriptive and self-explanatory
- Uses professional language
- Is specific rather than generic
For the summary:
- Summarize the key points in maximum two sentences
- Focus on what was discussed, decided, or accomplished
- Be concise but informative
<transcript>
{text}
</transcript>
"""
).strip()
class TopicResponse(BaseModel):
"""Structured response for topic detection"""