Shubhang S

A transactional safety layer for multi-agent LLM systems - deterministic replay, cryptographic provenance, verifier - gated consensus, and prefix rollback.

ThreatFind is a real-time autonomous network threat hunting system that combines Vision Transformers (ViT) for anomaly detection with a Retrieval-Augmented Generation (RAG) pipeline powered by Llama 3.1. The system is orchestrated as an agentic workflow and visualized through a live SOC-style dashboard.

A RAG system built around one invariant: the vector index should never silently drift from the source of truth. Most RAG systems re-index on a batch schedule, so a document can change in Postgres and the answer engine keeps citing stale content indefinitely, with no signal that anything is wrong.

A production RAG API that answers questions grounded in your own documents, built the way it would run in production: async FastAPI, PostgreSQL + pgvector (HNSW ANN), Redis cache, two-stage retrieval (vector search → cross-encoder rerank), containerised, deployed to Kubernetes via GitOps, and fully observable.