RAG Web Chatbot
A retrieval-augmented chatbot that answers questions grounded in the websites and YouTube videos it is given.
- ownership
- Independent project
- when
- Sep 2025
- org
- Independent
The film, as text: RAG Web Chatbot. A retrieval-augmented chatbot that answers questions grounded in the websites and YouTube videos it is given. Scrape or transcribe: Websites are scraped and YouTube videos are transcribed. MiniLM embeddings: Content is chunked and embedded with all-MiniLM-L6-v2. Separate collections: Each source gets its own collection, so unrelated content never mixes. Retrieve: A question searches with FAISS and pulls the most relevant passages. Answer: The answer is generated from the retrieved source context. 90%+ response accuracy. Independent project, Independent, Sep 2025.
Problem.
Answers drawn from mixed sources blur together. Isolating each source in its own collection keeps every answer anchored to where it came from.
How it works
- Source ingestion: accepts website and YouTube URLs, scraping page content with BeautifulSoup or transcribing the video.
- Embeddings: source content is converted to dense vector embeddings with Hugging Face all-MiniLM-L6-v2 for semantic search through FAISS.
- Per-source collections: each source is stored separately, so unrelated content never mixes at retrieval time.
- Grounded answers: semantic search with FAISS and LangChain answers from the retrieved source context using a Groq-hosted LLM, in a Streamlit chat interface.
Outcome
90%+
response accuracy
Stack
- Retrieval
- RAG
- FAISS
- Vector embeddings
- Hugging Face all-MiniLM-L6-v2
- Generation
- LangChain
- Groq API
- Ingestion
- BeautifulSoup
- YouTube transcription
- Application
- Python
- Streamlit