Driverthing/PLAN.md

Mileage Tracking Automation — Plan

How it works

1. Open the web app on your phone
2. Take 3 photos in sequence:
   • Photo 1 — odometer at the start of ride 1
   • Photo 2 — odometer at the end of ride 1 / start of ride 2
   • Photo 3 — odometer at the end of ride 2
3. Claude Vision API reads each number automatically
4. Two trip rows are calculated and saved to the Excel sheet
5. Download the sheet anytime

---

Stack

Layer	Technology
Backend	Rust + Axum
AI / OCR	Claude API (vision)
Spreadsheet	rust_xlsxwriter (Excel)
Frontend	Mobile-friendly HTML

---

Excel sheet columns

Date	Time	Odometer start (km)	Odometer end (km)	Trip (km)	Notes
2026-03-18	08:14	84 273	84 320	47	Ride 1
2026-03-18	09:05	84 320	84 391	71	Ride 2

---

Files to create

Driverthing/
├── src/
│   ├── main.rs         # Axum server + routes + session state
│   ├── claude.rs       # Claude API vision call
│   └── excel.rs        # Excel read/write logic
├── templates/
│   └── index.html      # Mobile step-by-step camera upload UI
├── Cargo.toml
└── mileage_log.xlsx    # Generated, gitignored

---

Session state (in-memory)

Between the 3 uploads the server holds a simple session with:

session {
  reading_1: Option<u32>   # start of ride 1
  reading_2: Option<u32>   # end of ride 1 / start of ride 2
  reading_3: Option<u32>   # end of ride 2
}

When all three readings are present, two rows are written to Excel and the session is cleared.

---

UI flow (single page, steps replace each other)

Step 1: "Take photo of odometer — START of ride 1"
        [Camera button] → upload
            ↓ server reads: 84 273 ✓

Step 2: "Take photo of odometer — END of ride 1 / START of ride 2"
        [Camera button] → upload
            ↓ server reads: 84 320 ✓
            ↓ Ride 1: 47 km (shown on screen)

Step 3: "Take photo of odometer — END of ride 2"
        [Camera button] → upload
            ↓ server reads: 84 391 ✓
            ↓ Ride 2: 71 km (shown on screen)
            ↓ Both rows saved ✓

Done screen: summary + [Download Excel] + [Start new session]

---

Flow (technical)

POST /upload?step=1  →  read photo → store reading_1 in session
POST /upload?step=2  →  read photo → store reading_2 → calc ride1 delta → show
POST /upload?step=3  →  read photo → store reading_3 → calc ride2 delta
                     →  write 2 rows to Excel → return summary
GET  /download       →  serve mileage_log.xlsx

---

Implementation steps

1. Set up Axum server with in-memory session state (DashMap or Mutex)
2. Create a single /upload endpoint that accepts step=1|2|3 + multipart image
3. Send each uploaded image (base64) to Claude Vision API: "What is the odometer reading in km? Reply with only the number."
4. Parse the integer from the response and store it in the session
5. After step 2: calculate ride 1 km, return intermediate confirmation
6. After step 3: calculate ride 2 km, append both rows to the Excel file, clear session
7. Mobile-friendly step-by-step HTML UI (progress indicator, camera capture)
8. GET /download endpoint to serve the Excel file