Takeoff AI

Takeoff AI is an AI application for automating the production of civil engineering takeoffs from PDF documents.

Users upload a plan PDF, preview it in the browser, and start a background extraction workflow that stores the file, queues processing, and sends a notification when the job is done.

                     PLAN PDF

                        │

         ┌──────────────┴──────────────┐

         │                             │

         ▼                             ▼

   Rendered pages                PDF metadata

   / high-res images             / text / coords

         │                             │

         ▼                             │

   Vision model                        │

         │                             │

         │  "This is a wall"           │

         │  "This is a door"           │

         │  "This is a dimension"      │

         │  "This room is bedroom"     │

         │                             │

         └──────────────┬──────────────┘

                        ▼

                STRUCTURED OBJECTS

                        │

                        ▼

             PYTHON GEOMETRY ENGINE

                        │

            ┌───────────┴───────────┐

            ▼                       ▼

       Measurements             Validation

            │                       │

            └───────────┬───────────┘

                        ▼

                 ANNOTATED PLAN

                        │

                        ▼

                      BEME

Stack

• Frontend: Next.js + TypeScript
• Backend: Express + TypeScript
• Worker: Python + BullMQ
• Preview: PDF.js
• Storage: shared Docker volume
• Queue: Redis
• Database: Postgres

Folder Structure

.
├── frontend/
│   ├── app/
│   │   ├── globals.css
│   │   ├── layout.tsx
│   │   └── page.tsx
│   ├── components/
│   │   ├── pdf-page.tsx
│   │   └── pdf-preview.tsx
│   ├── lib/
│   │   └── pdfjs.ts
│   ├── public/
│   ├── Dockerfile
│   └── package.json
├── server/
│   ├── src/
│   ├── worker/
│   ├── compose.yaml
│   ├── compose.dev.yaml
│   ├── Dockerfile
│   └── package.json
└── README.md

Workflow

1. The frontend uploads a PDF and renders the pages locally with PDF.js.
2. The Express API stores the file in the shared volume and creates a BullMQ job.
3. The Python worker processes the document and sends a notification when it finishes.
4. Postgres and Redis support the workflow behind the scenes.

Run

Development

docker compose -f server/compose.dev.yaml up --build

This starts the frontend, API, worker, Postgres, and Redis with development build targets and file watching.

Production

docker compose -f server/compose.yaml up --build

This uses the production targets for the frontend, API, and worker containers.

API Notes

See server/README.md for the backend endpoints and service-specific run instructions.