I build reliable software
that scales quietly.

Clinical software users were spending real time reviewing and summarizing patient conversations by hand. The friction held back engagement, and support was absorbing the slow turnaround.

Integrated Azure OpenAI / GPT-4 into the clinical product to generate conversation summaries directly inside the user’s workflow — one of the earliest production LLM deployments in a regulated healthcare context. Worked through the regulated-software constraints around data flow, evaluation, and safe rollout.

20% lift in user engagement and a 10% reduction in support response time — measurable proof that LLM-assisted workflow could be safely shipped in a regulated product.

An automated process could only run a single job at a time. As demand grew, the queue grew with it, and the only knob anyone could turn was “wait longer.”

Restructured the workflow so each job was independent and idempotent, moved coordination off the host, and added concurrency controls so we could safely run hundreds of jobs simultaneously without stepping on shared resources.

Throughput went up roughly two orders of magnitude, backlog dropped to near zero, and the team got back the engineering hours they were spending babysitting the queue.