Applied AI Engineer
K Health
I build and own production AI systems at K Health — autonomous agents, real-time anomaly detection, and LLM analytics platforms serving enterprise health system partners.
- Built Hawk-GPT, a production Slack bot serving 6 enterprise health system partners — an autonomous analytics agent that diagnoses operational problems through multi-step investigation (causal metric graph, sub-queries, root-cause synthesis). Combines hybrid RAG (semantic + BM25 over 794 schema fields), template matching, self-correcting query execution, and full eval/observability pipeline. Non-technical ops staff now self-serve analytics that previously required analyst support.
- Designed and deployed a real-time anomaly detection platform monitoring the full telehealth patient journey — video calls, intake, SMS, registration, surveys — across 6 partners. 15 statistical checks (z-score anomaly detection against 4-week seasonal baselines), LLM transcript classification with session recovery detection, and a calibrated AI overseer validated via backtesting against 34+ historical scenarios. Runs every 15 minutes; cut mean time-to-detection from hours to minutes.
- Built and deployed an autonomous AI agent for ops teams — a config-driven, two-pass LLM platform (Gemini) that analyzes call transcripts for patient sentiment, intent, and resolution patterns. Applied across 13+ campaigns and 4 partners. First analysis identified that 23.7% of inbound volume had the lowest resolution rate, directly reshaping the product roadmap.
- Engineered 15+ Claude Code skills as multi-agent workflows — git branching, validation, metrics comparison, MR generation, and ticket updates — compressing multi-hour processes into single commands.