Rafael B.

I build AI systems for healthcare and life-sciences teams that need clinical accuracy, an audit trail, and a human in the loop. Most AI developers in this space come from software and learn the domain on your dime. I come from the domain. I hold a Master's in Bioengineering, worked in pharmaceutical manufacturing and operations at GSK, Iovance, and Amneal, and now work in pharma commercial analytics running SQL-heavy data workflows in a regulated environment. I know why validation, review, and traceability are not optional in health data, because I have worked inside the systems that enforce them. What I build for healthcare teams: - Multi-agent clinical pipelines that take an intake or assessment, run it through reasoning and deterministic validation, and produce a personalized output a clinician reviews before it ships. - HIPAA-aware backends that let an existing clinical app call an LLM safely, with patient data separated, a BAA-covered model provider, and every call logged. - RAG assistants grounded in your own protocols and documents, with source citations on every answer. - Evaluation and guardrail layers that measure whether output is actually correct, not just plausible. Recent results: - Built and run a multi-agent clinical reporting system in production for Tom Sult, MD. It scores patient assessments against his own encoded decision rules, drafts a personalized report, and holds it in a clinician approve, edit, or reject queue before anything reaches the patient. He approved thirteen of thirteen unified reports with no edit or rejection. One silent failure hid a member for forty nine days, so I built the detection that would have caught it on day one. It runs behind more than 840 automated tests, eight static analysis checks, and seven scheduled monitors including a daily synthetic end-to-end canary. - Cut a researcher's manual data analysis workload by 50% with a custom AI tool. - Built a support agent that resolves 70% of inbound inquiries autonomously. That first one is the difference. Plenty of people can get an LLM to produce something that looks right in a demo. Far fewer build the validation, the failure detection, and the audit trail that let a clinician trust the output six months later. You get systems that hold up in production, not demos that fall apart on real data. Tech I work in: Python, FastAPI, PostgreSQL, Supabase, Railway, Langfuse, n8n, plus the Anthropic and OpenAI APIs. I also build outside healthcare. If you have a workflow eating hours every week, the same engineering applies with a different domain on top: AI agents that complete multi-step tasks end to end rather than single prompts, RAG assistants over your own documents, chat and support agents that resolve routine inquiries, and n8n automation that connects your tools and moves data between them. A note on compliance: I design HIPAA-aware systems with the appropriate controls, meaning access control, encryption, audit logging, and approval gates. I do not represent any system as certified compliant. Formal compliance is confirmed with your counsel and your BAAs. Send me a short note about the task and the outcome you want. I will tell you honestly whether AI is the right tool and how I would build it.