Ben Drolet
I help healthcare startups take AI from pilot to production, compliantly.
I help healthcare startups get their AI out of pilot and into production, on HIPAA-ready infrastructure with real evals and observability, so their engineers can ship AI features safely and fast.
The Problem
Most healthcare organizations trying to build with AI hit the same wall: the product vision is clear, but the infrastructure underneath isn't.
Taking an AI pilot to production in healthcare is harder than it looks. LLM deployments break without the right abstractions. Engineers rebuild the same tooling on every project. HIPAA and regulatory constraints add requirements that generic AI architectures don't handle.
Getting the infrastructure right early determines whether your AI program ships or stalls.
What I Do
LLM Platform Engineering
Design and build the infrastructure for LLM-powered applications — model routing, prompt management, evaluation pipelines, LLM observability, hallucination detection, and deployment systems that survive model upgrades.
Internal AI Developer Platforms
Build the tooling and abstractions that let product engineers use AI without becoming AI infra engineers — shared components, standardized access patterns, AI guardrails, and the HIPAA-compliance layer that healthcare requires.
Selected Work
From duct tape to a production AI platform
Remote-care healthcare company
The AI pilots worked, but nobody could trust them in production. Prompts were hard-coded into services, evals were done by hand, and there was no single place to enforce PHI handling or see what was failing.
I led the platform build:
- —A central prompt service with versioning and tracing
- —An evaluation pipeline that catches regressions before a prompt or model change ships
- —One path for every model call, so access control, PHI boundaries, audit logging, and model routing live in one place
- —AWS infrastructure in Terraform, with dashboards and alerting on failure rates, fallbacks, and latency
A monitored pilot was running within two weeks. Changes that used to take days took hours: when a product manager asked one morning to add a patient's heart-failure risk to a prompt's output, it was live in production after lunch, through CI and approvals, with PHI flowing through it compliantly.
The same platform went on to carry a per-patient snapshot, then a clinician chat agent that answers questions about a patient's history from charts, specialist-visit transcripts, and lab results.
About
16 years as a backend software engineer. For the past 4 years, focused on healthcare — behavioral health, digital health, and health tech — where the infrastructure problems are harder and the cost of getting it wrong is higher. Recently focused on the infrastructure side of AI — building the LLM platforms, evaluation pipelines, and internal developer tooling that production AI applications run on.
I work with healthcare engineering teams that have serious AI ambitions and need the infrastructure layer built right.
Based in San Francisco.
Contact
If you're building AI infrastructure in healthcare, I'd like to hear what you're working on.
[email protected]