DoktorDedi
Doctor said → done.
Doctor Said — so you don't forget.
Doctor's note → confirmed plan → reminders → answers from your own note.
Problem
The doctor explains. The patient forgets at the door.
The questions come at home, at night.
The visit ends; what the doctor said stays with the patient.
Live demo · You are the doctor
“Metformin twice a day, after meals; check-up in one month; if you get chest pain, go to the ER.”
Türkçe: “Metformin günde iki, yemekten sonra; bir ay sonra kontrol; göğüs ağrısı olursa acil.”
Then the patient asks: “Can I take it on an empty stomach?”
What you just saw
AI assistantDoctor's note → a care plan you confirm
Dose reminders at the times you choose (sample voice)
Your question answered from your confirmed note, with the source, in the avatar's voice
Trust
No source, no answer.
- Voiceexpo-audio
- StructureOpenAI Structured Outputs
- Confirmpatient / relative
- Scheduledeterministic code
- Avatar answerthe patient's note + leaflet only
It finds and reads what the doctor said and the TİTCK patient leaflet (Kullanma Talimatı). It never diagnoses and never changes a dose. Red flag → one answer: 112.
Built with Codex and tools
One spec, parallel Codex agents, one PR per task.
Codex · OpenAI API · fal · Expo
Evidence
Where the numbers come from.
- #1 Public repo + CI gate
- #2 Mobile avatar chat · 17/17 tests, CI run 38222650986
- #3 Server chat + voice APIs · 18/18 tests, CI run 38222647078 · TR/EN transcription 2/2 (TR 3.3 s, EN 2.0 s; Metformin, twice / after meals preserved), manual live check, src/server/VALIDATION.md
- #4 Release checks
- Spec: AGENTS.md + TODO
- Parallel Codex agents, one worktree each
- One PR per task
- CI gate
- Codex + Claude juries merge
Codex (agents, PRs, reviews) · OpenAI API (Structured Outputs, gpt-transcribe) · fal (avatar, TTS, voice-clone API) · Expo (iOS, Android, web)
fal, measured: face 21.3 s · TTS 2.4–28.5 s · lip-sync 15.2 min → the app uses portrait + voice. Claude: site, deck, docs.
Customer value
Reminder apps are everywhere.
We bring the doctor's words home.
Customer: hospitals
Post-discharge follow-up.
Reminders work
SMS reminders: adherence 36.7% → 96.7% (Turkish RCT).
Ekinci et al., Konuralp Tıp Derg 2017AI is where health money goes
54% of US digital-health funding went to AI in 2025.
Rock Health, Jan 2026Next: hospital HBYS · e-Nabız integration
- ProblemThe doctor explains, the patient forgets; the questions come at home.
- It worksYou saw it: care plan you confirm → dose reminders → sourced answer in the avatar's voice.
- Codex + tools3 merged Codex PRs behind a CI gate · OpenAI Structured Outputs · fal avatar + voice · Expo
- CustomerHospitals' post-discharge follow-up · 9.6M adults with diabetes in Türkiye
- Doctor said → done.The visit ends; what the doctor said stays with the patient.
doktordedi.aemreusta.com