solo-developer

Autonomous agent for email management

GPT-4o-realtime APIGemini 2.5 ProGoogle's MedLM/MedGemmacustom prompting on Manchester triage protocolsynthetic scenarios from historical recordsfeedback collection UIadmin panel
Stack tools7
AddedMar 2026
StatusPublished

Deployed in Mexican hospitals within 30 days full-time work; achieved better than human triage accuracy; ongoing testing by doctors and academics for wider licensing; patients prefer AI over waiting for staff; 15% accuracy/UX improvement via backend oversight.

solo-developer

Why they built it

Requested by a doctor to handle patient flow during staff shortages in emergency wards/clinics where patients wait too long for triage, enabling faster prioritization.

What worked

Out-of-the-box Gemini 2.5 Pro performed well on triage especially subtle cases humans miss; patients overwhelmingly choose quick AI over staff wait times; backend LLM quorum improved reliability by 15%; rapid prototyping led to hospital deployment.

What broke or was painful

Google's live APIs poor quality and lack of emotiveness requiring Frankenstein architecture with frontend/backend split; real-world noise cancellation issues; EHR integration delayed by hospital IT responsiveness; constant manual feedback review (5-10 hrs/week); diminishing returns on accuracy improvements.

The result

Deployed in Mexican hospitals within 30 days full-time work; achieved better than human triage accuracy; ongoing testing by doctors and academics for wider licensing; patients prefer AI over waiting for staff; 15% accuracy/UX improvement via backend oversight.

References