I sit at the intersection of business operations and applied AI engineering. Most of my work starts as a business problem — an SLA at risk, a fraud/violation review process that's too slow, a finance dataset nobody trusts an LLM near — and ends as a system that makes deterministic, auditable decisions in production, not just LLM suggestions.
Business pain point → Constraint-aware system design → Shipped, measurable outcome
| Business problem | System | Outcome |
|---|---|---|
| Traffic violation review is slow & inconsistent | GridLock-VioScan YOLOv8 + Gemini Vision detection, automated severity scoring (VSI) & fine-waiver logic | Evidence-backed, auditable violation pipeline |
| SRE teams need to trust AI triage before it touches prod | IncidentOps-Agentic-SRE OpenEnv benchmark scoring LLM triage / diagnosis / remediation | Measurable trust bar before agentic response ships |
| Finance can't risk LLM hallucination on O2C data | Dodge-AI-FDE SAP order-to-cash data modeled as a graph; LLM is a strict router, never a guesser | Zero-hallucination query layer over sensitive financial data |
| Farm ops need decisions committed, not advised | AI-Autonomous-Farming-Operator Deterministic safety pipeline + conversational multi-agent layer (EN/HI) | System authorizes & verifies real actions, not recommendations |
| Support orgs need SLOs held under load | supportops-env OpenEnv benchmark: AI coordinator triaging tickets & holding SLOs | Reproducible stress-test for agentic support ops |
| Hidden system failures need diagnosis, not dashboards | sentinel-os-agentic-benchmark Trajectory-scored diagnostic reasoning benchmark | Quantified recovery efficiency for LLM agents |
