Podcasts to Knowledge

Converts YouTube podcast/interview sessions into a structured knowledge graph with CocoIndex

🌟 Please help star CocoIndex if you like this project!

Pipeline:

1. Read YouTube URLs from plain text files
2. Download audio via yt-dlp, transcribe with speaker diarization via AssemblyAI
3. Extract metadata, speakers, and thematic statements via LLM (openai/gpt-5.4-mini)
4. Resolve duplicate entities (persons, techs, orgs) using CocoIndex's entity_resolution utility (embedding similarity via faiss + LLM confirmation via pydantic-ai)
5. Store the knowledge graph in SurrealDB: sessions, statements, persons, techs, orgs, and their relationships

Prerequisites

• Python 3.11+
• FFmpeg installed (required by yt-dlp for audio extraction)
• Docker (for SurrealDB)
• An AssemblyAI API key (for transcription with speaker diarization)
• An OpenAI API key (for LLM extraction via litellm)

Setup

1. Start SurrealDB

Run SurrealDB with persistent storage via Docker:

docker run -d \
  --name surrealdb \
  --user root \
  -p 8787:8000 \
  -v surrealdb-data:/data \
  surrealdb/surrealdb:latest \
  start --user root --pass root surrealkv:/data/database

This persists data in a Docker volume (surrealdb-data) across container restarts.

2. Set environment variables

# Required
export ASSEMBLYAI_API_KEY="..."
export OPENAI_API_KEY="sk-..."

# Optional (shown with defaults)
export SURREALDB_URL="ws://localhost:8787/rpc"
export SURREALDB_NS="cocoindex"
export SURREALDB_DB="yt_conversations"
export SURREALDB_USER="root"
export SURREALDB_PASS="root"
export INPUT_DIR="./input"
export LLM_MODEL="openai/gpt-5.4-mini"
export RESOLUTION_LLM_MODEL="openai/gpt-5-mini"

3. Install dependencies

pip install -e .

4. Add YouTube URLs

Edit input/sample.txt (or create new .txt files under input/). One URL per line, # for comments:

# AI podcasts
https://www.youtube.com/watch?v=VIDEO_ID_1
https://www.youtube.com/watch?v=VIDEO_ID_2

Run

Build/update the knowledge graph:

cocoindex update conv_knowledge.app

This is incremental — re-running skips sessions that haven't changed.

Visualize the Knowledge Graph

SurrealDB includes a built-in web UI called Surrealist for exploring and visualizing data.

Option A: Surrealist Cloud

1. Go to app.surrealdb.com
2. Connect to your local instance:
   • Endpoint: ws://localhost:8787
   • Namespace: cocoindex
   • Database: yt_conversations
   • Username: root / Password: root
3. Use the Explorer tab to browse tables and relations
4. Use the Query tab to run SurrealQL queries (see below)

Option B: Surrealist Desktop

Download from surrealdb.com/surrealist for a native app with the same features.

Example queries

See all relationships:

SELECT * FROM statement_involves, session_statement, person_session;

Browse all sessions and their statements:

SELECT id, name, description, date FROM session;

Find all statements a person made:

SELECT
  <-person_statement<-person.name AS speaker,
  statement
FROM statement;

Explore the full graph around a person:

SELECT
  name,
  ->person_session->session.name AS sessions,
  ->person_statement->statement.statement AS statements
FROM person;

Find all entities involved in a statement:

SELECT
  statement,
  ->statement_involves->person.name AS persons,
  ->statement_involves->tech.name AS techs,
  ->statement_involves->org.name AS orgs
FROM statement;

Top N techs mentioned by the most people:

SELECT
  name,
  array::distinct(
    <-statement_involves<-statement<-person_statement<-person.name
  ) AS persons,
  array::len(array::distinct(
    <-statement_involves<-statement<-person_statement<-person.id
  )) AS person_count
FROM tech
ORDER BY person_count DESC
LIMIT 10;

Schema

Nodes:  session, statement, person, tech, org
Edges:  session_statement  (session -> statement)
        person_session     (person -> session)
        person_statement   (person -> statement)
        statement_involves (statement -> person | tech | org)