🎙️ YouTube Transcription & Chat Agent

A fast, offline-friendly proof of concept (POC) to transcribe YouTube videos and chat with the content using a local LLM via Ollama. Built with Flask, Whisper, PostgreSQL, and Mistral.

---

✅ Features

• 🧠 Transcribe YouTube audio using Whisper for Transcribe operation and Mistral Model for Chat
• 💬 Chat with transcript using Ollama (Mistral, LLaMA2, etc.)
• 📦 Local PostgreSQL-backed storage
• ⚙️ Configurable via .env

---

🔧 Prerequisites

• Python 3.10+
• Docker
• Ollama with mistral:7b-instruct-fp16 model

---

🚀 Getting Started

1. Clone the Project

git clone https://github.com/13shivam/yt-agent.git
cd yt-agent

2. Make sure a .env File exists

FLASK_APP=app.py
FLASK_ENV=development

# Ollama setup
OLLAMA_MODEL=mistral:7b-instruct-fp16
OLLAMA_API=http://localhost:11434/api/chat

# PostgreSQL connection
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_DB=ytagent
POSTGRES_USER=postgres
POSTGRES_PASSWORD=postgers

# Whisper model
WHISPER_MODEL=base

3. Run Local Ollama with Mistral Model (7b-instruct-fp16 - 14gb)

brew install ollama
ollama run mistral:7b-instruct-fp16

---

4. Run App

docker compose build --no-cache
docker compose up -d

---

Access APIs via Swagger

http://127.0.0.1:5050/apidocs/#

---

Sequence Diagram all APIs

Swagger API Local Demo

Step 1: Upload API youtube URL

Step 2: Check JobStatus

Step 3: Initiate Chat from JobId

Misc Screenshots

1. DB Screenshots
2. Chat API preview
3. Flask App Logs

🪪 WIP to add in future release:

• Speaker Diarization support via open source pyannote.audio
• Speaker Diarization — NVIDIA NeMo Framework

📝 License

This project is licensed under the MIT License - see the license file for details.

Important Notice Regarding Open Source Dependencies:

This project relies on various open-source models and libraries, each with its own licensing terms. It is the user's responsibility to understand and adhere to the specific licenses of all the open-source components they choose to use. Consult the individual licenses provided by the respective model and library providers.