Agentic AI Engineer — I build multi-agent systems: orchestrators, sub-agents, routing, hooks and guardrails.
Lead Software Engineer at Persistent Systems · previously enterprise RAG and agents at Nomura.
I design agentic systems that hold up in production — the orchestration, delegation and governance around LLMs, not just the prompt.
- 🕸️ Multi-agent orchestration — a single orchestrator breaking work into bounded tasks and delegating to specialist sub-agents in a hierarchical tree, on GitHub Copilot custom agents.
- 🔀 Routing & hand-offs — per-caller allow-lists of child agents (no global registry),
path-scoped instructions, and a structured status contract (
completed / partial / blocked / failed) so the orchestrator aggregates and escalates deterministically. - 🛡️ Guardrails — least-privilege tool scoping, human-in-the-loop approval gates, lifecycle hooks, MCP tool integration and agent memory.
- 🤖 Agent Mode at Nomura — multi-agent orchestration in LangGraph with routing, state and hand-offs, and MCP tool-calling into Confluence, Jira and financial data platforms.
- 🔎 Enterprise RAG for 15,000+ users — hybrid retrieval (BM25 + dense on PGVector, Reciprocal Rank Fusion, cross-encoder re-ranking) and a RAG evaluation pipeline (faithfulness, answer relevancy, contextual precision/recall) so changes are measured, not guessed.
- 🔧 Sonar AutoFix — an agentic CI/CD remediation platform that cleared ~40% of SonarQube issues across enterprise microservices and won a Nomura Star award.
- Designing Agent Hierarchies in VS Code Copilot — why one agent with every tool is a liability, and how a hierarchy of narrow agents (orchestrators, layered context, least-privilege tools, human approval) fixes it. Code: Agentic-Orchestration-Harness.
- AI Engineering Roadmap: A Free Playbook for Builders
| Project | What it is |
|---|---|
| Agentic-Orchestration-Harness | Hybrid orchestration framework for VS Code GitHub Copilot custom agents — orchestrator → sub-agent hierarchy, routing without a global registry, structured status contracts, least-privilege tools, approval gates and optional hooks/MCP/memory. Installs into any repo's .github/. |
| ai-engineering-playbook ⭐63 | A zero-to-100 learning path for applied AI engineering — RAG, embeddings, vector search, agents, MCP and the production engineering around them. 56 pages, built as a searchable site. |
| CodeRAG | Code-aware RAG that models source code by structure — method signatures, call graphs, static analysis — so retrieval follows real code relationships. Python · Sentence Transformers · ChromaDB/PGVector/Weaviate · React · TypeScript |
| TCS-NQT-CODING-SHEET ⭐985 · Infosys-SP-Coding-Questions ⭐148 | Curated solution sets for the TCS NQT and Infosys SP hiring tests. ~1,100 stars combined. |
Agentic & LLM engineering
Retrieval & data
Languages & infra
B.Tech, Electronics & Communication Engineering — VIT Vellore (2020 – 2024), CGPA 8.71/10.