personal-health-portal

Raw STATUS.md

# personal-health-portal

**Status:** 🟢 Active  
**Phase:** Discovery  
**Last Activity:** 2026-02-06 (Normalized STATUS.md format)
**Created:** 2026-02-04  
**Owner:** John Girard

---

## Linear Metadata

**Project ID:** `cc044667-96fa-41da-a9db-9a12baefdb21`
**Team ID:** `96b685fe-2252-47c5-97ee-273d8c484942`
**Last Synced:** `2026-02-06T13:21:57.992Z`

## Vision

Web application that ingests lab results and medical imaging, uses AI to parse and structure the data, logs it in a database, visualizes changes over time, and provides personalized coaching and feedback based on trends.

---

## Requirements (Defined 2026-02-04)

### Product Vision
✅ **Primary user:** Just you (personal tool, not a product)  
✅ **Core value prop:** Lab results with graphs of change over time + personal LLM-driven analysis/advice - "Chat with my labs"  
⏳ **Comparison:** TBD - research existing tools for UI/feature inspiration

### Scope & Features
✅ **Data sources:** PDF uploads (lab results) for now  
✅ **AI parsing:** BOTH extraction AND interpretation (flag abnormals, discuss them)  
✅ **Visualization:** Time-series charts (definite), others TBD as we build  
✅ **Coaching/feedback:** AI-generated recommendations  
✅ **Sharing:** Might share reports (with Amber, doctors)

### Technical
✅ **Privacy:** Either local or cloud is fine  
✅ **Timeline:** Exploratory (no rush)  
⏳ **Tech stack:** Recommendations needed  
⏳ **Budget/constraints:** TBD

---

## Tech Stack Recommendation

### Frontend
**Next.js 14+ (App Router) + Tailwind CSS + shadcn/ui**
- **Why:** Modern, fast, great DX. Can deploy to Vercel or run locally. shadcn gives you polished UI components out of the box.
- **Chart library:** Recharts (React-friendly, good for time-series)

### Backend & Database
**Option A: Supabase (Postgres + Auth + Storage)**
- **Pros:** Managed Postgres, built-in auth (if you want to share later), file storage for PDFs, generous free tier
- **Cons:** Cloud dependency (but you said either is fine)

**Option B: Local SQLite + Drizzle ORM**
- **Pros:** No cloud dependency, fast, simple, portable (single file database)
- **Cons:** Can't easily share; manual backups

**Recommendation:** Start with **Supabase** - you can always export/migrate later, and it makes sharing easier if you want to show Amber.

### AI/LLM
**Anthropic Claude (Sonnet 4.5 or Opus 4.5)**
- **Why:** Best at medical/scientific reasoning, structured outputs, long context (good for full lab reports)
- **Alternative:** OpenAI GPT-4 (cheaper, faster, but Claude is better at nuanced health discussions)

### PDF Parsing
**pdf-parse (npm package) or LlamaParse (specialized for complex PDFs)**
- **Why:** Extract text from lab PDFs, then use Claude to structure the data
- **Workflow:** Upload PDF → extract text → Claude parses into structured JSON → store in DB

### Architecture
```
User uploads PDF
  ↓
PDF text extraction (pdf-parse)
  ↓
Claude API: "Extract lab values from this report"
  ↓
Structured JSON (date, test name, value, unit, reference range)
  ↓
Store in Supabase (lab_results table)
  ↓
Display in time-series chart (Recharts)
  ↓
"Chat with labs" → RAG over stored results + Claude
```

---

## MVP Scope

## Notes


Created 2026-02-04 based on initial description:
> "web app that takes in lab and imaging results and uses AI to parse them, log them in a table. then can show changes over time visually and provide coaching / feedback"