feat: add background information for room context in summaries

server/migrations/versions/082fa608201c_add_room_background_information.py

@@ -0,0 +1,26 @@
+"""add_room_background_information
+
+Revision ID: 082fa608201c
+Revises: b7df9609542c
+Create Date: 2025-07-29 01:41:37.912195
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '082fa608201c'
+down_revision: Union[str, None] = 'b7df9609542c'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column('room', sa.Column('background_information', sa.Text(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column('room', 'background_information')

server/reflector/db/rooms.py

@@ -39,6 +39,7 @@
     sqlalchemy.Column(
         "is_shared", sqlalchemy.Boolean, nullable=False, server_default=false()
     ),
+    sqlalchemy.Column("background_information", sqlalchemy.Text),
     sqlalchemy.Index("idx_room_is_shared", "is_shared"),
 )
 
@@ -58,6 +59,7 @@ class Room(BaseModel):
         "none", "prompt", "automatic", "automatic-2nd-participant"
     ] = "automatic-2nd-participant"
     is_shared: bool = False
+    background_information: str = ""
 
 
 class RoomController:
@@ -106,6 +108,7 @@ async def add(
         recording_type: str,
         recording_trigger: str,
         is_shared: bool,
+        background_information: str = "",
     ):
         """
         Add a new room
@@ -121,6 +124,7 @@ async def add(
             recording_type=recording_type,
             recording_trigger=recording_trigger,
             is_shared=is_shared,
+            background_information=background_information,
         )
         query = rooms.insert().values(**room.model_dump())
         try:

server/reflector/pipelines/main_live_pipeline.py

@@ -454,15 +454,47 @@ class PipelineMainFinalSummaries(PipelineMainFromTopics):
     Generate summaries from the topics
     """
 
+    async def get_room(self):
+        """Get room information for the transcript"""
+        if not self._transcript.room_id:
+            return None
+        return await rooms_controller.get_by_id(self._transcript.room_id)
+
     def get_processors(self) -> list:
         return [
             TranscriptFinalSummaryProcessor.as_threaded(
                 transcript=self._transcript,
+                room=getattr(self, '_room', None),
                 callback=self.on_long_summary,
                 on_short_summary=self.on_short_summary,
             ),
         ]
 
+    async def create(self) -> Pipeline:
+        self.prepare()
+
+        # get transcript
+        self._transcript = transcript = await self.get_transcript()
+
+        # get room information
+        self._room = await self.get_room()
+
+        # create pipeline
+        processors = self.get_processors()
+        pipeline = Pipeline(*processors)
+        pipeline.options = self
+        pipeline.logger.bind(transcript_id=transcript.id)
+        pipeline.logger.info(f"{self.__class__.__name__} pipeline created")
+
+        # push topics
+        topics = self.get_transcript_topics(transcript)
+        for topic in topics:
+            await self.push(topic)
+
+        await self.flush()
+
+        return pipeline
+
 
 class PipelineMainWaveform(PipelineMainFromTopics):
     """

server/reflector/processors/summary/summary_builder.py

@@ -135,7 +135,7 @@ def print_content(self, role, content):
 
 
 class SummaryBuilder:
-    def __init__(self, llm, filename: str | None = None, logger=None):
+    def __init__(self, llm, filename: str | None = None, logger=None, room=None):
         self.transcript: str | None = None
         self.recap: str | None = None
         self.summaries: list[dict] = []
@@ -147,6 +147,7 @@ def __init__(self, llm, filename: str | None = None, logger=None):
         self.llm_instance: LLM = llm
         self.model_name: str = llm.model_name
         self.logger = logger or structlog.get_logger()
+        self.room = room
         self.m = Messages(model_name=self.model_name, logger=self.logger)
         if filename:
             self.read_transcript_from_file(filename)
@@ -465,26 +466,31 @@ async def generate_summary(self, only_subjects=False):
         self.logger.debug("--- extract main subjects")
 
         m = Messages(model_name=self.model_name, logger=self.logger)
-        m.add_system(
-            (
-                "You are an advanced transcription summarization assistant."
-                "Your task is to summarize discussions by focusing only on the main ideas contributed by participants."
-                # Prevent generating another transcription
-                "Exclude direct quotes and unnecessary details."
-                # Do not mention others participants just because they didn't contributed
-                "Only include participant names if they actively contribute to the subject."
-                # Prevent generation of summary with "no others participants contributed" etc
-                "Keep summaries concise and focused on main subjects without adding conclusions such as 'no other participant contributed'. "
-                # Avoid: In the discussion, they talked about...
-                "Do not include contextual preface. "
-                # Prevention to have too long summary
-                "Summary should fit in a single paragraph. "
-                # Using other pronouns that the participants or the group
-                'Mention the participants or the group using "they".'
-                # Avoid finishing the summary with "No conclusions were added by the summarizer"
-                "Do not mention conclusion if there is no conclusion"
-            )
+
+        system_prompt = (
+            "You are an advanced transcription summarization assistant."
+            "Your task is to summarize discussions by focusing only on the main ideas contributed by participants."
+            # Prevent generating another transcription
+            "Exclude direct quotes and unnecessary details."
+            # Do not mention others participants just because they didn't contributed
+            "Only include participant names if they actively contribute to the subject."
+            # Prevent generation of summary with "no others participants contributed" etc
+            "Keep summaries concise and focused on main subjects without adding conclusions such as 'no other participant contributed'. "
+            # Avoid: In the discussion, they talked about...
+            "Do not include contextual preface. "
+            # Prevention to have too long summary
+            "Summary should fit in a single paragraph. "
+            # Using other pronouns that the participants or the group
+            'Mention the participants or the group using "they".'
+            # Avoid finishing the summary with "No conclusions were added by the summarizer"
+            "Do not mention conclusion if there is no conclusion"
         )
+
+        # Add room context if available
+        if self.room and self.room.background_information:
+            system_prompt += f"\n\nContext about this meeting room: {self.room.background_information}"
+
+        m.add_system(system_prompt)
         m.add_user(
             f"# Transcript\n\n{self.transcript}\n\n"
             + (

server/reflector/processors/transcript_final_summary.py

@@ -12,9 +12,10 @@ class TranscriptFinalSummaryProcessor(Processor):
     INPUT_TYPE = TitleSummary
     OUTPUT_TYPE = FinalLongSummary
 
-    def __init__(self, transcript=None, **kwargs):
+    def __init__(self, transcript=None, room=None, **kwargs):
         super().__init__(**kwargs)
         self.transcript = transcript
+        self.room = room
         self.chunks: list[TitleSummary] = []
         self.llm = LLM.get_instance(model_name="NousResearch/Hermes-3-Llama-3.1-8B")
         self.builder = None
@@ -23,7 +24,7 @@ async def _push(self, data: TitleSummary):
         self.chunks.append(data)
 
     async def get_summary_builder(self, text) -> SummaryBuilder:
-        builder = SummaryBuilder(self.llm)
+        builder = SummaryBuilder(self.llm, room=self.room)
         builder.set_transcript(text)
         await builder.identify_participants()
         await builder.generate_summary()

server/reflector/views/rooms.py

@@ -33,6 +33,7 @@ class Room(BaseModel):
     recording_type: str
     recording_trigger: str
     is_shared: bool
+    background_information: str
 
 
 class Meeting(BaseModel):
@@ -55,6 +56,7 @@ class CreateRoom(BaseModel):
     recording_type: str
     recording_trigger: str
     is_shared: bool
+    background_information: str = ""
 
 
 class UpdateRoom(BaseModel):
@@ -67,6 +69,7 @@ class UpdateRoom(BaseModel):
     recording_type: str
     recording_trigger: str
     is_shared: bool
+    background_information: str = ""
 
 
 class DeletionStatus(BaseModel):
@@ -108,6 +111,7 @@ async def rooms_create(
         recording_type=room.recording_type,
         recording_trigger=room.recording_trigger,
         is_shared=room.is_shared,
+        background_information=room.background_information,
     )
 
 

www/app/(app)/rooms/page.tsx

@@ -11,13 +11,14 @@ import {
   Input,
   Select,
   Spinner,
+  Textarea,
   createListCollection,
   useDisclosure,
 } from "@chakra-ui/react";
 import { useEffect, useState } from "react";
 import useApi from "../../lib/useApi";
 import useRoomList from "./useRoomList";
-import { ApiError, Room } from "../../api";
+import { Room } from "../../api";
 import { RoomList } from "./_components/RoomList";
 
 interface SelectOption {
@@ -54,6 +55,7 @@ const roomInitialState = {
   recordingType: "cloud",
   recordingTrigger: "automatic-2nd-participant",
   isShared: false,
+  backgroundInformation: "",
 };
 
 export default function RoomsList() {
@@ -170,6 +172,7 @@ export default function RoomsList() {
         recording_type: room.recordingType,
         recording_trigger: room.recordingTrigger,
         is_shared: room.isShared,
+        background_information: room.backgroundInformation,
       };
 
       if (isEditing) {
@@ -189,11 +192,10 @@ export default function RoomsList() {
       setNameError("");
       refetch();
       onClose();
-    } catch (err) {
+    } catch (err: any) {
       if (
-        err instanceof ApiError &&
         err.status === 400 &&
-        (err.body as any).detail == "Room name is not unique"
+        err.body?.detail === "Room name is not unique"
       ) {
         setNameError(
           "This room name is already taken. Please choose a different name.",
@@ -215,6 +217,7 @@ export default function RoomsList() {
       recordingType: roomData.recording_type,
       recordingTrigger: roomData.recording_trigger,
       isShared: roomData.is_shared,
+      backgroundInformation: roomData.background_information || "",
     });
     setEditRoomId(roomId);
     setIsEditing(true);
@@ -323,6 +326,20 @@ export default function RoomsList() {
                 {nameError && <Field.ErrorText>{nameError}</Field.ErrorText>}
               </Field.Root>
 
+              <Field.Root mt={4}>
+                <Field.Label>Background Information</Field.Label>
+                <Textarea
+                  name="backgroundInformation"
+                  placeholder="Provide context about this room's purpose, meeting type, or any relevant background information that will help generate better summaries..."
+                  value={room.backgroundInformation}
+                  onChange={handleRoomChange}
+                  rows={3}
+                />
+                <Field.HelperText>
+                  This information will be used to provide context for AI-generated summaries
+                </Field.HelperText>
+              </Field.Root>
+
               <Field.Root mt={4}>
                 <Checkbox.Root
                   name="isLocked"

www/app/(app)/rooms/useRoomList.tsx

@@ -1,18 +1,18 @@
 import { useEffect, useState } from "react";
 import { useError } from "../../(errors)/errorContext";
 import useApi from "../../lib/useApi";
-import { Page_Room_ } from "../../api";
+import { PageRoom } from "../../api";
 
 type RoomList = {
-  response: Page_Room_ | null;
+  response: PageRoom | null;
   loading: boolean;
   error: Error | null;
   refetch: () => void;
 };
 
 //always protected
 const useRoomList = (page: number): RoomList => {
-  const [response, setResponse] = useState<Page_Room_ | null>(null);
+  const [response, setResponse] = useState<PageRoom | null>(null);
   const [loading, setLoading] = useState<boolean>(true);
   const [error, setErrorState] = useState<Error | null>(null);
   const { setError } = useError();

www/app/(app)/transcripts/useTranscriptList.ts

@@ -1,10 +1,10 @@
 import { useEffect, useState } from "react";
 import { useError } from "../../(errors)/errorContext";
 import useApi from "../../lib/useApi";
-import { Page_GetTranscriptMinimal_, SourceKind } from "../../api";
+import { PageGetTranscriptMinimal, SourceKind } from "../../api";
 
 type TranscriptList = {
-  response: Page_GetTranscriptMinimal_ | null;
+  response: PageGetTranscriptMinimal | null;
   loading: boolean;
   error: Error | null;
   refetch: () => void;
@@ -16,7 +16,7 @@ const useTranscriptList = (
   roomId: string | null,
   searchTerm: string | null,
 ): TranscriptList => {
-  const [response, setResponse] = useState<Page_GetTranscriptMinimal_ | null>(
+  const [response, setResponse] = useState<PageGetTranscriptMinimal | null>(
     null,
   );
   const [loading, setLoading] = useState<boolean>(true);