SnapFuel

Phase 1 scaffolding for importing historical fuel-stop photos, grouping related images, and capturing fuel-stop details through a manual web workflow.

Solution layout

• FuelImport.Core – domain models, options, image metadata extraction, and grouping services for manual review.
• FuelImport.Data – EF Core FuelImportDbContext schema + migrations for source images, grouped fuel events, and review history.
• FuelImport.Aws – legacy AWS OCR integration kept only as a reference during the transition away from OCR.
• FuelImport.Web – manual-entry UI, folder scanning endpoint, export API endpoints, and single database management.
• FuelImport.Tests – focused unit/integration tests for pairing, confidence, validation, and dedup idempotency.

Run locally

dotnet restore SnapFuel.slnx
dotnet build SnapFuel.slnx
dotnet test FuelImport.Tests/FuelImport.Tests.csproj

Manual Entry & Photo Ingestion Web App

Configure FuelImport.Web/appsettings.json:

• Import:RootFolder to your photo root path (e.g. C:\Users\steph\Downloads\SnapFuelPhotos)
• Import:DryRun to scan without persisting source images

Run:

dotnet run --project FuelImport.Web

Open the web app in your browser (e.g., http://localhost:5000 or https://localhost:7001).

Click Scan Photos in the header toolbar to scan your configured photo folder into the database on-demand.

Endpoints:

• POST /api/import/scan
• GET /api/import/config
• GET /api/events?needsReview=true|false
• GET /api/events/{id}
• POST /api/events/{id}/review
• GET /api/manual/groups
• POST /api/manual/groups/split
• POST /api/manual/groups/merge
• POST /api/manual/groups/save
• GET /api/images/{id}
• GET /api/vehicles
• GET /api/events/report
• GET /api/events/log
• GET /vehicles.html
• GET /report.html
• GET /log.html
• GET /api/events/export/csv

Open the web UI to work through grouped image sets. Each group is built from nearby timestamps and matching GPS coordinates from EXIF when available. Saving a group creates or updates one fuel event linked to every image in that group.

Notes

• This implementation is intentionally practical for personal use.
• OCR is no longer part of the primary workflow; manual entry is now the source of truth.
• It does not require training custom ML models.