Fix Ai Hallucinations & Automate Scheduling
Publicada el 2026-07-13
Descripción de la oferta
I rely on GPT-3 inside my clinical practice software to streamline patient interactions and pre-visit paperwork, yet the model sometimes invents facts and produces clinically unsafe text. I need your help to eliminate these “hallucinations” and to design a rock-solid appointment-scheduling flow that ties seamlessly into our existing tools. Here’s what I’m trying to achieve: • Stabilise GPT-3 responses so they never output clinically inaccurate information again. A mix of prompt engineering, post-processing filters, external knowledge look-ups, or light fine-tuning is acceptable—the key is measurable accuracy. • Build an automated scheduling pipeline that pulls requests from our web form, checks current availability, confirms with the patient, and writes back to the EHR and calendar without human intervention. Right now the breakage occurs at the calendar step, creating double-bookings. Acceptance criteria 1. A benchmark script demonstrating <1% factual error on a 100-item test set of typical patient queries. 2. A live demo where a patient books, reschedules, and cancels an appointment with no clashes in Google Calendar and our EHR sandbox. 3. Clear documentation and hand-off instructions so my in-house staff can maintain the flow. I can provide API keys, anonymised data, and access to our current Zapier + Node.js glue code. If you have experience placing guardrails around GPT-3 and wiring up healthcare calendars through HIPAA-friendly tooling, let’s talk.
Skills
Fuente original: freelancer