DocMind AI is a privacy-first Retrieval-Augmented Generation (RAG) system designed to bridge the gap between secure local intelligence and high-performance cloud LLMs. It allows users to chat with massive PDF datasets using a Hybrid Inference Bridge that preserves data sovereignty while providing citation-backed accuracy.
Launch the platform in 2 steps:
# 1. Start Backend (API + Vector DB)
docker-compose up -d --build
# 2. Start Frontend Dashboard
cd frontend && npm install && npm run devDetailed Setup: See GETTING_STARTED.md.
High-fidelity chat UI with real-time neural indexing telemetry.
Hybrid Inference Gateway routing between OpenAI (Cloud) and Ollama (Local).
Deep observability into the vector store and semantic document chunks.
Deep Dive: See ARCHITECTURE.md for Chunking Logic and Decision Logs.
- 🧠 Hybrid Brain: Switch between GPT-4o and Llama 3 instantly.
- 📚 RAG Pipeline: Professional recursive splitting (1000/200 overlap).
- 🔍 High-Precision Search: Hybrid semantic + metadata filtering.
- 🔒 Air-Gapped Ready: Fully local vector storage using ChromaDB.
Understanding how a PDF becomes a conversational agent:
- Ingest: Document parsed and cleaned via
pypdf. - Chunk: Segmented into 1000-char overlapping blocks.
- Embed: Converted to high-dimensional vectors.
- Index: Stored in ChromaDB with page-level metadata.
- Query: System retrieves top chunks to ground LLM responses.
| Document | Description |
|---|---|
| System Architecture | Vectors, Chunking, and Provider Abstraction. |
| Getting Started | Enviroment setup (Cloud vs Local mode). |
| Failure Scenarios | Hallucination mitigation and grounding logic. |
| Interview Q&A | RAG strategy and technical justifications. |
| Component | Technology | Role |
|---|---|---|
| Brain | FastAPI (Python) | LangChain Orchestrator. |
| Memory | ChromaDB | Local Vector Store. |
| Intelligence | OpenAI / Ollama | LLM Inference Backends. |
| Interface | Next.js 14 | Enterprise Dashboard. |
Harshan Aiyappa
Senior Full-Stack Hybrid Engineer
GitHub Profile
This project is licensed under the MIT License - see the LICENSE file for details.