Mentor Chatbot (RAG)
Engineered a local, privacy-preserving Retrieval-Augmented Generation (RAG) chatbot to query an extensive internal knowledge base of over 200 institutional documents. The system solves LLM hallucination and knowledge cutoffs by acting as an 'Open Book' assistant.
System Architecture
- Chunking Strategy: 500 characters with 50-character overlap using RecursiveCharacterTextSplitter.
- Embeddings: Hugging Face sentence-transformers/all-MiniLM-L6-v2 (384-dimensional).
- Vector Store: 4,680 embedded chunks stored persistently in ChromaDB.
- Retrieval: k=10 similarity search feeding into a local Llama 3.1 8b instance via Ollama.
PythonStreamlitLangChainChromaDBOllamaLlama 3.1
.png&w=96&q=75){kind=small}
