νοῦς — the mind that reads, reasons, and writes
A multimodal RAG platform for research teams. Ingest papers and documents, query them with a LangGraph agent, and synthesise answers backed by a knowledge graph and vector search.
39 screenshots in the full gallery
Capabilities
Every layer of the stack is purpose-built for retrieval quality, not demo impressiveness.
LangGraph Agent
StateGraph with intent routing across research, writing, data, and general subgraphs. Twelve specialised tools, 30 s timeout, max 3 concurrent via semaphore.
Hybrid Search
Vector embeddings (Qdrant) and knowledge-graph traversal (Neo4j) combined with deterministic ranking. Answers cite real sources, not hallucinations.
Human-in-the-Loop
Destructive tools (ingest, create_note, create_draft) trigger interrupt(). The client polls, reviews, and resumes with explicit confirmation.
arXiv Integration
Search, ingest, and cite papers directly inside the agent conversation. Full-text extraction, entity linking, and graph indexing happen automatically.
Supabase SSR Auth
Cookie sessions, no client-side token storage. Email verification, forgot-password, and reset flows. IDOR-hardened with validated enums throughout.
GitOps Deployment
DigitalOcean Kubernetes with ArgoCD and Helm across dev / staging / prod. Self-hosted CI runner with Docker layer cache on DO Spaces. LangSmith + Prometheus observability.
History
What shipped, and when.
April 2026
March 2026