Working read · 11 Aug 2026
What the files are — and what to build
Combined take from WhatsApp, discharge samples, evaluation form, and (offline) historical sheets.
What’s in the folder
| File | What it really is |
|---|---|
document.txt | WhatsApp group: product asks (screening → matching) |
3 .rtf files | Max CLBS discharge summaries (sample data), not chats |
evaluation-forms/ | Signature Hospital liver recipient checklist photos |
excel-data/ | Confidential historical donor/recipient workbooks — local only |
One family case:
- Nasibakhon — DCLD HCV → dual-lobe LDLT recipient
- Mukhriddin — son, left-lobe donor
- Tuichivai — brother, left-lobe donor
Matches Selva’s ask: give Manu discharge summaries and build an AI follow-up model.
What they want
One liver-clinic product in layers:
- At registration — quick screening for all patients.
- Sheet-like registry — add every new patient/donor in structured fields (replace Excel chaos).
- Match — hard rules (GRWR > 0.65, remnant ≥ 30% @ ~10% steatosis, age < 75, no major comorbidities) then optional complication foresight.
- After diagnosis / discharge — pathways by type: liver/DCLD; post-transplant; living donor.
- Behind it — registry + later AI; keep simple; patient → app interface.
See Matching. HEAL Group / Facebook clips = inspiration only.
What to build
CLBS sheet + match + follow-up assistant — not a general hospital AI, not auto-prescribing ChatGPT.
v1
- Screening questionnaire at registration
- Google-Sheet-like recipient + donor intake (evaluation checklist fields)
- Hard-rule match calculator with clear pass/fail reasons
- Doctor-approved advice templates (confirm before patient sees)
- Basic registry row from screening + match + discharge fields
v2
- Complication pattern hints from historical columns (never publish raw rows)
- WhatsApp/SMS reminders + DS extraction
- Light AI: summarise, flag missed follow-up — not invent treatment
Do not start with full autonomous AI prescribing or “rejected case” ML (no reject data). HLA is legal, not medical match. Decisions in India: legal → financial → medical.
Mismatch to resolve with Selva
Chat DCLD pack ≠ post-LT recipient pack. Donors and recipients already differ. Need pathway templates, not one medicine list for everyone. Also need the explicit “major comorbidity” list for auto-fail.
Next build: sheet intake → hard match → screening form → advice template → registry row.