Skip to content

PROJECT_PROGRESS.md

Latest commit

 

History

History
127 lines (107 loc) · 4.36 KB

PROJECT_PROGRESS.md

File metadata and controls

127 lines (107 loc) · 4.36 KB
Raw

PROJECT_PROGRESS.md

Current milestone

Final refactor from thickness/CML extraction to isometric drawing to Material Take-Off (MTO) generator.

Completed tasks

  • Replaced thickness/CML domain models with MTO domain models.
  • Added DrawingMeta, MTOItem, MTOSummary, and MTO-focused ExtractionResult.
  • Added strict category/unit enums:
    • PIPE, FITTING, FLANGE, VALVE, GASKET, BOLT, SUPPORT
    • M, EA, NO, SET
  • Updated mock pipeline to return a realistic sample MTO.
  • Updated Gemini prompt to extract Material Take-Off from piping isometric drawings.
  • Added JSON schema-style response contract in gemini_vision.py.
  • Kept automatic mock fallback when GEMINI_API_KEY is missing or live extraction fails.
  • Updated post-processing to:
    • normalize pipe rows;
    • compute pipe length totals;
    • count fittings/flanges/valves/supports;
    • derive gasket rows;
    • derive bolt-set rows;
    • compute summary cards.
  • Updated CSV export to output MTO fields.
  • Updated storage history metadata from point count/job details to line number/drawing number/item count.
  • Updated frontend types to match backend MTO schema.
  • Updated upload flow with client-side validation and local preview URL.
  • Updated results UI to show:
    • drawing preview;
    • drawing metadata;
    • MTO summary cards;
    • MTO table;
    • assumptions/notes/warnings;
    • CSV download.
  • Updated history UI to show MTO status and row counts.
  • Updated README with architecture, setup, AI pipeline, assumptions, limitations and trade-offs.
  • Removed generated frontend TypeScript build cache from final ZIP preparation.
  • Backend test suite passes locally: 18 passed.

Remaining tasks

  • Run frontend npm install and npm run typecheck on a machine with npm dependencies available.
  • Test live Gemini extraction using a real GEMINI_API_KEY.
  • Take screenshots of the running app if required by the evaluator.
  • Optionally deploy frontend/backend and add hosted URLs to README if submitting a hosted demo.

Folder structure

project/
├── frontend/
│   ├── app/
│   ├── components/
│   ├── lib/
│   └── types/
├── backend/
│   ├── app/
│   │   ├── core/
│   │   ├── models/
│   │   ├── routers/
│   │   ├── services/
│   │   └── utils/
│   ├── sample_assets/
│   └── tests/
├── README.md
├── PROJECT_PROGRESS.md
├── .env.example
└── .gitignore

Current APIs

Method Endpoint Status
GET /api/health Implemented
POST /api/upload Implemented
POST /api/extract Implemented
GET /api/export/{file_id}.csv Implemented
GET /api/history Implemented
GET /api/results/{file_id} Implemented

Implemented features

  • Next.js App Router frontend.
  • FastAPI backend.
  • PDF/image upload.
  • Server-side validation.
  • PDF/image preprocessing.
  • Gemini Vision extraction path.
  • Mock pipeline fallback.
  • Pydantic validation.
  • MTO post-processing.
  • Gasket/bolt derivation.
  • Summary cards.
  • CSV export.
  • SQLite or in-memory storage.
  • Swagger docs.
  • Backend tests.

Pending features / possible bonus work

  • Docker Compose.
  • Excel export.
  • Bounding-box overlay.
  • Async job queue.
  • Multi-sheet rollup.
  • OCR/BOM table extraction.
  • Manual correction workflow.

Known issues

  • Live Gemini accuracy is unverified in this environment because no real API key was used.
  • Frontend typecheck was not executed here because dependencies may not be installed in the runtime.
  • Cached results produced by the old thickness/CML schema are incompatible with the new MTO schema. Delete backend/data/ if reusing an old local run.

Design decisions

  • Kept existing upload -> extract -> export endpoint pattern to minimize risk and preserve the working project shell.
  • Used synchronous extraction because the assessment scope is one drawing per upload.
  • Used SQLite as default storage so local history persists across restarts.
  • Used mock fallback as a first-class feature for evaluator reproducibility.
  • Derived gaskets and bolt sets in post-processing because they are often not explicitly drawn but are required MTO consumables.

Next milestone

Run the application locally, test with the sample drawing, optionally test with a real Gemini key, capture screenshots, and prepare final submission ZIP.