I build failure-aware backend systems and grounded AI workflows, then connect them to focused product experiences. My work emphasizes durable state, clear service boundaries, measurable quality, and observable operations.

A full-stack checkout workflow designed for duplicate delivery, concurrency, and partial failure—not just the happy path.

• Reliable events: transactional outbox, bounded retries, dead-letter handling, and idempotent consumers
• Consistent workflow: atomic inventory reservation, optimistic locking, and compensation when payment fails
• Operational feedback: health probes, Prometheus metrics, correlation IDs, integration tests, and full-topology CI

Repository · Architecture · ADRs · Testing · Release v0.1.0

I design AI systems that are grounded, evaluated, and deployable.

• Grounded retrieval: conversation-aware PDF question answering with hybrid FAISS + BM25 retrieval, rank fusion, citation validation, and evidence checks
• Structured generation: source-grounded quiz generation with constrained outputs, validation, duplicate rejection, and explainable ranking
• Multimodal delivery: infrared and visible-light image fusion behind FastAPI, with durable cloud jobs and autoscaling deployment

Selected implementations are currently maintained in private repositories.

Current direction: safe agentic commerce using scoped tools, approval gates, audit logs, and regression evals for product discovery and order operations.

detekt/detekt · Merged contribution clarifying multi-preview annotation suppression in the Kotlin static-analysis documentation.

Backend, data, and platform

AI and product

LinkedIn is the best place to reach me.