Holusight
Holusight is a local-first document intelligence engine. Point it at any folder of documents and it builds a hybrid search index — combining BM25 keyword search (SQLite FTS5) with vector embeddings (LanceDB) merged via Reciprocal Rank Fusion. Ask it a question, get a synthesized answer with source citations.
The Problem
Enterprise teams drown in documents. Critical knowledge is locked in PDFs, Word files, and wikis. Standard search returns results — not answers. LLMs hallucinate without grounding. Existing tools require cloud upload and expensive SaaS contracts.
The Solution
A local Python library + web UI that runs entirely on your machine. Index any folder in seconds, search with hybrid BM25 + vector retrieval, and get LLM-synthesized answers grounded in your actual documents. Pluggable LLM backends: Claude, OpenAI, Azure, or Ollama for fully offline use.
The Outcome
Document search goes from keyword guessing to natural language Q&A. Runs locally — no data leaves the machine. Pluggable into any stack via Python API, CLI, or Streamlit chat UI.
Key Features
- Hybrid BM25 + vector search merged with Reciprocal Rank Fusion
- LLM answer synthesis grounded in document chunks (no hallucination)
- Pluggable LLM backends — Claude, OpenAI, Azure OpenAI, Ollama (offline)
- Local-first — all data stays on your machine
- Python API + CLI + Streamlit web chat UI
- Supports PDF, Word, Markdown, plain text
Technology Stack
Interested in this project?
I'd love to discuss the technical details, challenges overcome, or similar projects I could build for you.
View Live AppLet's discuss this project