01AI· 2026
WisdomLens
Retrieval-augmented answers grounded in a real knowledge base.
A chat-based knowledge assistant that answers questions from a stored text knowledge base. Semantic search over 800+ chunks with Sentence-Transformers + FAISS finds the most relevant passages, and a RAG pipeline feeds that context to a locally-run Llama 3.2 (via Ollama) for grounded, bilingual answers.
- Problem
- LLMs answer confidently even when they don't know — and can't cite where an answer came from. WisdomLens keeps answers tied to real source passages.
- Engineering challenge
- Keeping retrieval, generation and the chat UI as independent parts behind a REST API, so each could evolve without breaking the others — and running the model locally through Ollama instead of a paid API.
- Status
- Working project. FastAPI backend + React frontend; retrieval over 800+ text chunks with bilingual responses.
PythonFastAPIReactFAISSSentence-TransformersOllamaLlama 3.2
Repository link not listed
SYSTEM FLOW
Knowledge base
Source text enters the system.
Chunking
Split into 800+ semantic chunks.
Embeddings
Each chunk becomes a vector.
Vector search
FAISS nearest neighbours to the query.
Retrieval
Top-k relevant context selected.
Llama 3.2
Answer generated, grounded in context.
Answer
Traceable, source-linked response.