R&D • Artificial Intelligence
JARVIS
A personal, multi-model AI orchestrator with vector memory.
Node.js / LangGraph / RAG / R&D
The need
A personal R&D project, with no client brief, exploring LLM orchestration, long-term memory and voice/vision interfaces.
The solution
A central orchestrator (LangGraph) routes requests across several model providers (Anthropic, Google Gemini, Groq, OpenAI) depending on the task.
A two-tier memory system: persistent SQLite storage and a Qdrant vector database, with RAG-style indexing and similarity retrieval.
Independent voice (wake-word detection, Whisper transcription, speech synthesis) and vision (screen capture, OCR, image analysis) pipelines, decoupled from the orchestration core.
The architecture is documented — the key decisions (overall architecture, cognitive model, memory system) are written up; the rest of the planned documentation is still in progress.
- Multi-model orchestration (LangGraph)
- Persistent + vector memory (SQLite, Qdrant) with RAG
- Voice pipeline: wake word, transcription, speech synthesis
- Vision pipeline: screen capture, OCR, image analysis
- Application-level permissions / security portal
Gallery
Result
A working, documented codebase, frozen as a stable base ahead of a new iteration currently in private development.