I build multi-agent systems and production-oriented LLM applications, with experience in telecom log intelligence, refinery decision optimization, crude procurement, and legal AI.

My focus is the engineering around models: workflow orchestration, retrieval, optimization, evaluation, observability, safety boundaries, and human-in-the-loop decision making.

Auditable multi-agent incident triage for synthetic telecom logs. It combines log-template normalization, anomaly evidence, topology reasoning, runbook matching, a critic agent, and approval-gated remediation.

Python · Multi-Agent · AIOps · FastAPI · Docker

Explainable crude-procurement and refinery-planning demo. Mathematical optimization produces the plan while specialized agents surface assumptions, supply risks, constraint checks, and decision rationale.

Python · Optimization · Decision Intelligence · Supply Chain · Agents

Citation-first legal research workflow using fictional materials. It demonstrates retrieval, grounded drafting, citation verification, abstention, and mandatory professional review.

Python · RAG · Legal AI · Responsible AI · Evaluation

• Agents should have bounded roles, typed state, observable decisions, and measurable outcomes.
• LLMs explain and coordinate; deterministic code and optimization enforce hard constraints.
• Enterprise AI needs evaluation, access control, auditability, and human approval by design.
• Public demonstrations should use synthetic or openly licensed data, never confidential material.

Multi-agent orchestration · Agent evaluation · RAG quality · LLM observability · MCP · AI-assisted operations research · Private and local model deployment

The featured repositories are independent portfolio reconstructions built with public or synthetic data. They do not contain proprietary code, customer data, or internal materials from any employer or client.