Senior AI Engineer · Physicist by training · I build production LLM and agentic systems.

I'm an AI engineer architecting the orchestration, tracing, and evaluation stack of a production multi-agent platform, covering RAG pipelines, tool calling, guardrails, gold-label datasets, and CI release gates. Five years of production AI sit behind that: agentic systems, LLM fine-tuning and deployment, retrieval, and observability for generative AI, plus earlier work on fraud detection at millions of transactions a day.

I did my PhD in physics, building unsupervised deep neural networks for quantum many-body systems, including a new approach to the Fermion Sign Problem. I still think the most interesting problems live where rigorous mathematics meets messy production engineering.

• Agentic AI: multi-agent orchestration (pydantic-graph, LangGraph), tool calling, human-in-the-loop
• Eval & observability: tracing, scorer harnesses, gold-label datasets, CI release gates
• RAG & retrieval: ingestion, retrieval, and response evaluation over enterprise knowledge bases
• LLM fine-tuning: managed fine-tuning on Azure OpenAI, QLoRA / LoRA / PEFT for domain tasks
• Computer-use & automation: perception, planning, and action loops driving legacy systems with no API
• The physics → AI bridge: equivariant networks, the Fermion Sign Problem, deep learning for quantum systems

Agentic AI

• Sutra: a graph linking code, decisions, and sessions that Claude Code can query
• Shopping chat agent: a conversational commerce agent

Enterprise LLM & RAG

• RAGKYC: retrieval-augmented KYC and document verification
• Retail Insights Assistant: an LLM assistant for retail analytics
• Intelligent Accountant: an LLM-driven bookkeeping assistant

Physics → AI

• Deep-learning quantum simulations: unsupervised deep learning for quantum many-body systems

Languages

GenAI & agents

LLMs & fine-tuning

ML / deep learning

Data & MLOps

Cloud

Currently exploring how far you can push agentic systems before the abstractions leak.