• 💼 AI Engineer @ Innoflexion Technologies — building production agentic AI & data platforms
• 🔭 Currently building Nexus MCP — a temporal-knowledge-graph PM platform for AI coding agents
• 🌱 Deepening: agentic system design, knowledge graphs, and backend infra at scale
• 🏆 Smart India Hackathon (SIH) National Finalist — Top 6 nationwide, Grand Finale at IIT Hyderabad
• 👯 Open to collaborating on interesting agentic/RAG/dev-tooling projects
• 📫 Reach me at work.bhanu2004@gmail.com

AI Engineer @ Innoflexion Technologies · Jun 2025 – Present

• Designed and built a conversational analytics AI agent (natural language → SQL) that queries live relational databases (PostgreSQL, MSSQL, Oracle) and flat-file/Parquet datasets through a single unified agent interface, built on the Agno agent framework.
• Implemented long-term agent memory as a temporal knowledge graph (Neo4j + Graphiti), including migrating the embedding layer to AWS Bedrock (Titan) embeddings.
• Built a semantic query cache that recognizes previously-answered natural language questions and short-circuits redundant LLM/DB round-trips, reducing latency and cost for repeated query patterns.
• Built an OCR + document ingestion pipeline (Mistral OCR) with hybrid full-text + vector (pgvector) retrieval for unstructured documents and email content.
• Built a multi-source data connector handling database, flat-file, and mailbox ingestion, orchestrated with Apache Airflow, landing on a shared Parquet-based data layer.
• Deployed and operated backend services with FastAPI, Docker, and AWS Fargate; used Redis for caching/session state; exposed agent tools to external clients via FastMCP (Model Context Protocol).
• Stack: Python · FastAPI · PostgreSQL + pgvector · Neo4j · Redis · Apache Airflow · Docker · AWS (Bedrock, Fargate) · Agno · FastMCP

Org-wide Agentic PM & Temporal Knowledge Graph Platform for AI Coding Agents (Antigravity, Claude Code, Cursor, Codex)

Every org, project, and developer gets a role-scoped AI chatbot backed by a living, commit-verified temporal knowledge graph — fed automatically by coding agents at session end, so teams stop paying twice for the same work.

• 🧠 LLM-free direct Cypher ingestion — deterministic, no hallucinated graph writes
• 🗄️ Dual-DB storage: PostgreSQL 16 + pgvector (relational/vector) and Neo4j AuraDB (temporal knowledge graph)
• ✅ Ground-truth verification — knowledge graph claims are checked against real Git commit history, not agent self-reporting
• 🔐 Multi-layer RBAC — parameterized tenant-rooted Cypher traversals + OPA/Rego policy engine, row-level security via tenant-scoped Postgres sessions
• 🔌 MCP-native — FastMCP SSE server with JWT-secured handshake, tested live against Claude Desktop and Cursor's MCP client
• 📦 One-command onboarding — npx @nexusmcp/cli init scaffolds .mcp.json with credential shielding
• ☁️ Deploying to Azure; currently running on Neo4j AuraDB + containerized PostgreSQL

Stack: FastAPI · PostgreSQL + pgvector · Neo4j · FastMCP · Redis · OPA/Rego · React · Docker · Azure