Case Study
YouTube Growth AI Chatbot (Genzzz)
RAG chatbot over YouTube transcripts. FAISS retrieval feeds Gemini 2.5 Flash, deployed free on Hugging Face Spaces.
Executive Summary
Genzzz is a RAG chatbot over YouTube creator transcripts, giving conversational access to growth strategies and channel insights. Built in 7 days and deployed free on Hugging Face Spaces with Gemini 2.5 Flash.
Problem & Constraints
YouTube growth knowledge is buried across hours of video. The system retrieves the relevant transcript segments and generates accurate, context-aware answers.
Architecture
YT-DLP + Whisper transcripts → LangChain RecursiveCharacterTextSplitter → sentence-transformers embeddings → FAISS IndexFlatL2 → LangChain retrieval chain → Gemini 2.5 Flash → Gradio ChatInterface on HF Spaces.
Methodology
- Downloaded and cleaned YouTube transcripts (YT-DLP + Whisper fallback)
- Chunked at 512-1024 chars with RecursiveCharacterTextSplitter
- Embedded with all-MiniLM-L6-v2; built FAISS vector index
- Deployed Gradio UI on Hugging Face with GOOGLE_API_KEY secret
- Validated retrieval with natural language growth/strategy queries
Results & Metrics
| Metric | Result |
|---|---|
| Retrieval | FAISS L2 search |
| LLM | Gemini 2.5 Flash |
| Deploy | HF Spaces (free) |
| Demo | Hugging Face Space |
Tech Stack
Python, LangChain, FAISS, Gemini API, Gradio, Hugging Face Spaces, sentence-transformers
Future Work
Multi-channel support, real-time analytics integration, fine-tuned embedding models.