I design production-grade AI for regulated industries.

What I build

• Enterprise AI — systems that operate inside real enterprise constraints.
• Agent systems — orchestration, tools, MCP, RAG, validation.
• Reliable AI — security, governance, observability, deterministic control.

How I think about production AI

• Deterministic control.
• Least privilege.
• Validation before trust.
• Human control for high-impact decisions.

Selected work

• Eliza4J
  Java library for enterprise AI agent integration with BNY Eliza AI Agents and LLMs.
• FINOS DTCC India AI Hackathon
  3rd Prize
  Agentic onboarding-to-invest workflow · 40+ organizations (CTRL+ALT+GEEKS, BNY).
• DataLake Wealth
  Reusable MongoDB data-access layer for wealth platform services.

More projects →

Recent thinking

• Stop letting your LLM see your secrets
  Production AI Patterns · #17 · 2026-07-29

  Agentic workflows need bidirectional PII masking through a continuous protection gateway, not one-time front-door filtering.

• Stop asking your LLM to behave
  Production AI Patterns · #16 · 2026-04-18

  Security must be a decoupled AI Firewall—not a system prompt asking the model to behave.

• Multi-agent systems need a Human Gateway
  Production AI Patterns · #15 · 2026-04-04

  High-impact multi-agent decisions need an explicit Human Gateway for approval before execution.

• The God Agent is dead. Long live multi-agent routing
  Production AI Patterns · #14 · 2026-03-25

  Enterprise scale needs specialized agents and routing—not one omniscient God Agent.

All writing →