Job Description

Founding AI Engineer (Computer Vision)

Location: Remote

Employment Type: Full-time, Permanent

Salary Range: competitive

Start Date: ASAP

The Opportunity

We are a high-growth PropTech start-up building the next generation of estimation software for the UK construction industry. We are developing an automated "Takeoff & Cost Planning" solution that uses Artificial Intelligence to instantly detect rooms, walls, and materials from PDF construction drawings.

We are seeking a Founding AI Engineer to lead the technical architecture of this product. You will be building a "Raster-to-Vector" engine from the ground up, creating a tool that rivals established platforms like Kreo, CostX, or Bluebeam.

This is a hands-on, architect-level role. You will not just be tuning models; you will be building the intellectual property that drives our entire business.

The Challenge

Construction drawings are complex. They are often heavy PDF files containing scanned data, hand-written notes, and differing scales (1:100, 1:50). Your objective is to build an intelligence layer that can:

1. Ingest & Process: Handle large-scale floor plans using tiling strategies.

2. Visualise: Accurately segment walls, windows, and doors with pixel-level precision.

3. Vectorise: Convert "blobs" of pixels into clean, mathematically precise vector polygons for accurate quantity surveying.

Key Responsibilities

? Pipeline Architecture: Design and build the end-to-end Python pipeline--from PDF ingestion to inference and post-processing.

? Model Training: Train and fine-tune State-of-the-Art Semantic Segmentation models (e.g., DeepLabV3+, U-Net) on floor plan datasets (e.g., CubiCasa5k).

? Geometric Engineering: Write robust algorithms using OpenCV and Shapely to clean up AI output. You will tackle challenges such as contour detection, polygon approximation (Ramer-Douglas-Peucker), and geometric simplification.

? API Development: Wrap your models in a high-performance FastAPI service to serve real-time predictions to our web frontend.

? Infrastructure: Manage cloud GPU resources (AWS/GCP) to ensure the system is cost-efficient and scalable.

Tech Stack

? Core: Python 3.10+, NumPy, Pandas.

? AI/Deep Learning: PyTorch (preferred) or TensorFlow.

? Computer Vision: OpenCV (cv2), Scikit-Image, Albumentations.

? Math/Geometry: Shapely, GeoPandas, NetworkX.

? Backend: FastAPI, PostgreSQL, Docker, AWS (S3/EC2).

Requirements

? Experience: 3-5+ years of commercial experience in Computer Vision, Deep Learning, or Data Science engineering.

? Geometric Logic: You must have a strong grasp of spatial mathematics. Recognising a wall is straightforward; calculating its area (m2) accurately based on pixel coordinates and scale ratios is the complex challenge we need you to solve.

? Code Quality: You write clean, production-ready Python code, not just Jupyter Notebook experiments.

? Right to Work: You must possess the legal right to work in the UK (we cannot offer visa sponsorship at this stage).

Nice to Have

? Background in Construction Tech, CAD, or GIS (Geographic Information Systems).

? Experience with OCR (Optical Character Recognition) for reading room tags and dimensions.

? Familiarity with frontend tech (React.js/Konva.js) is a bonus, helping you understand how your data is rendered for the user.

How to Apply

To demonstrate you have the specific technical depth we need, please include a brief answer (2-4 sentences) to the following question in your application:

"You are given a binary segmentation mask of a floor plan generated by a U-Net model. The wall edges are jagged and pixelated. Which specific libraries and algorithmic steps would you use to convert this mask into a clean set of vector coordinates suitable for a CAD layer?"

Job Types: Full-time, Permanent

Work Location: Remote