Build Meeting Notes Knowledge Graph from Google Drive — FalkorDB (CocoIndex v1)

Extract structured information from meeting notes stored in Google Drive and build a knowledge graph in FalkorDB. The flow ingests Markdown notes, splits them by headings into per-meeting sections, uses an LLM (via LiteLLM + instructor) to parse participants, organizer, time, and tasks, and writes nodes and relationships into the graph.

Please drop CocoIndex on Github a star to support us and stay tuned for more updates. Thank you so much 🥥🤗.

What this builds

• Meeting nodes — one per meeting section, keyed by a stable integer id derived from (note_file, date)
• Person nodes — canonical organizers, participants, and task assignees, deduplicated by an embedding + LLM entity-resolution pass (so "Alice", "Alice Chen", and "alice c." collapse to a single node)
• Task nodes — tasks decided in meetings (keyed by description)
• Relationships:
  • ATTENDED — Person → Meeting (with is_organizer flag)
  • DECIDED — Meeting → Task
  • ASSIGNED_TO — Person → Task

The source is one or more Google Drive folders shared with a service account. The flow watches for changes and keeps the graph up to date incrementally.

How it works

The pipeline runs in three phases:

1. Per-file extraction. Read each file from Google Drive, split it by Markdown headings (# / ##) into meeting sections, and for each section extract a structured Meeting via LiteLLM + instructor (date, note, organizer, participants, tasks with assignees). Meeting and Task nodes plus DECIDED edges are declared in this phase. Raw person names are carried forward.
2. Person entity resolution. All raw person names from all files are deduplicated using sentence-transformer embeddings and an LLM pair resolver to produce a canonical-name mapping.
3. Person-touching relations. Canonical Person nodes are declared, then ATTENDED and ASSIGNED_TO edges are wired up using resolved names.

CocoIndex reconciles changes incrementally — re-running after editing one note only re-processes the affected sections, and the resolution phase only re-runs when the set of raw names changes.

Prerequisites

• A running FalkorDB instance:
  shCopy

  docker run -d -p 6379:6379 -p 3000:3000 falkordb/falkordb:latest
  

  The browser UI is at http://localhost:3000.
• An LLM key (defaults to OpenAI; configure via LLM_MODEL for other providers — see LiteLLM providers).
• A Google Cloud service account with read access to the source folders, and the folder IDs you want to ingest. See Setup for Google Drive.

Environment

Set the following variables (copy .env.example to .env and fill in):

export OPENAI_API_KEY=sk-...
export GOOGLE_SERVICE_ACCOUNT_CREDENTIAL=/absolute/path/to/service_account.json
export GOOGLE_DRIVE_ROOT_FOLDER_IDS=folderId1,folderId2
export FALKORDB_URI=falkor://localhost:6379
export FALKORDB_GRAPH=meeting_notes
export LLM_MODEL=openai/gpt-5.4
export RESOLUTION_LLM_MODEL=openai/gpt-5-mini   # used for entity resolution

Then:

set -a && source .env && set +a

Run

Install dependencies:

uv pip install -e .

Build/update the graph:

cocoindex update main

Browse the knowledge graph

Open the FalkorDB browser at http://localhost:3000, select the meeting_notes graph, and run Cypher queries.

// All relationships
MATCH p=()-->() RETURN p

// Who attended which meetings (including organizer; one edge per attendee)
MATCH (p:Person)-[:ATTENDED]->(m:Meeting)
RETURN p.name, m.note_file, m.time, m.id

// Tasks decided in meetings
MATCH (m:Meeting)-[:DECIDED]->(t:Task)
RETURN m.note_file, m.time, t.description

// Task assignments
MATCH (p:Person)-[:ASSIGNED_TO]->(t:Task)
RETURN p.name, t.description

// Meetings someone organized
MATCH (p:Person)-[r:ATTENDED {is_organizer: true}]->(m:Meeting)
RETURN p.name, m.note_file, m.time

You can also use redis-cli:

redis-cli GRAPH.QUERY meeting_notes \
  "MATCH (p:Person)-[:ATTENDED]->(m:Meeting) RETURN p.name, m.note_file, m.time"