AI / RAG · Prototype
Codex One
Privacy-first local knowledge base with RAG using local LLMs and vector search.
Role: Individual project
Problem
Need private, offline access to knowledge base with semantic search and answers.
Approach
Local RAG pipeline: document ingestion, chunking, embeddings, vector search, LLM inference via Ollama.
Engineering decisions
- • 100% local execution for privacy
- • Modular pipeline for ingestion/retrieval/generation
Challenges & trade-offs
- • Balancing retrieval quality with local model constraints
Notes
- • Prototype; evaluation metrics not fully published.