Back to Portfolio

All Projects Archive

Production multi-agent workflows, self-correcting RAG engines, and autonomous tool-calling systems engineered for real-world reliability.

01
GroundedAI

Self-Correcting Multi-Source Agentic RAG

Eliminates LLM hallucinations across complex multi-source documents by combining hybrid RRF vector search with local NLI faithfulness verification (94.2% grounding precision).
LangGraphFastAPIpgvectorNLI DeBERTa v3Next.jsRedis
02
rzp Merchant

Autonomous Agentic Commerce Engine (MCP/ACP)

Implements MCP enabling external AI agents to autonomously discover products, optimize baskets, and execute Razorpay transactions under multi-tier security mandates.
Next.jsLangGraphGemini 2.5Razorpay SDKMCPpgvector
03
MailMind

Agentic Email Processing & HITL Engine

Automates inbox operations with intent classification (96.5% accuracy), RAG draft generation, and PostgreSQL-persisted Human-in-the-Loop approval checkpoints (88% operational time reduction).
LangGraphFastAPIResendGroqFAISSPostgreSQL