04ActiveJan 2025 – Present

Nexus

A second brain — RAG-powered knowledge management

Personal Project

95%
Retrieval Accuracy
100%
Runs Locally
Mistral
LLM
ChromaDB
Vector DB

The Concept

Nexus is a personal knowledge management system that makes all your saved content queryable. Upload PDFs, paste article URLs, drop in markdown notes — then ask questions across your entire knowledge base with natural language. It's like having a conversation with everything you've ever read.

How It Works

Documents are chunked using a hybrid strategy (semantic + fixed-size with overlap) and embedded using sentence-transformers. Vectors are stored in ChromaDB with rich metadata. At query time, a LangChain retrieval chain fetches the top-k relevant chunks, then Mistral (via Ollama) generates a grounded answer with source citations. The system runs fully locally — no data leaves your machine.

What Makes It Different

Most RAG demos use naive chunking and get poor results on long documents. Nexus uses a custom chunking strategy that respects document structure — headings, paragraphs, and sections stay together. This is why it achieves 95% retrieval accuracy on benchmarked test sets. It also supports multi-document reasoning: ask a question that spans 3 different PDFs and get a synthesized answer.

Privacy First

Everything runs locally. Ollama serves Mistral on your own hardware — no OpenAI API keys, no cloud calls, no data sent anywhere. This makes it viable for notes on sensitive topics, private research, or client documents.

RAG Pipeline

Ingest PDF/URL/MDChunk + EmbedChromaDB StoreQueryRetrieve k chunksMistral GenerateAnswer + Citations

Tech Stack

PythonChromaDBLangChainOllamaMistralFastAPIReactPostgreSQL