Machine Learning Engineer - Computer Vision (Part-time)
Job Description
TekPiq is building an AI system that inspects building façades from drone imagery. It detects cracks, spalling, rust, water damage and energy leaks, then places each defect on a 3D model with a real-world size and location. We're looking for a part-time Machine Learning Engineer to own the computer vision core: the models that turn high-resolution drone photos into reliable, measurable defect findings. You'll work closely with the founder and a full-stack engineer, with room to shape the technical direction from the ground up. The role is remote or hybrid, with a clear path to full-time as the product grows. Experience with drone infrastructure inspection or similar computer vision projects is a strong plus.
Responsibilities
- •Train and ship defect-detection models (instance and semantic segmentation) for cracks, spalling, corrosion, staining and sealant failure on façade imagery
- •Build the image pipeline for very large photos: tiling, per-tile inference, and stitching detections back together across tile seams
- •Curate and grow our training data, starting from public datasets and moving to labeling our own drone images, using SAM and similar tools to speed up annotation
- •Connect detections to real-world measurement by back-projecting 2D findings onto a 3D photogrammetry model, so each defect gets a size and location
- •Define and track model metrics, run error analysis, and drive down false positives such as shadows, dirt and joints that look like cracks
- •Package models for deployment and hand clean, structured defect output to the reporting and dashboard layer
Qualifications
- •Strong Python and hands-on deep learning with PyTorch
- •Practical experience with image segmentation or detection models (e.g. YOLO-seg, U-Net, Mask2Former, SegFormer) that you have trained yourself, not just used off the shelf
- •Comfort with the full model loop: dataset curation, training, evaluation, and iterating on real-world failure cases
- •Ability to work with large, messy, real-world image data and limited labels
- •Self-directed and able to deliver on a part-time schedule with minimal oversight
- •Nice to have: photogrammetry or structure-from-motion (COLMAP, OpenDroneMap, Pix4D, RealityCapture)
- •Nice to have: experience with Segment Anything (SAM) for labeling or detection
- •Nice to have: 3D or geospatial work, such as point clouds, meshes, camera poses, or viewers like Three.js or Cesium
- •Nice to have: thermal or infrared imagery analysis
- •Nice to have: drone or UAV imagery and flight-planning data
- •Nice to have: MLOps, including labeling pipelines, retraining workflows, model deployment and monitoring
- •Nice to have: background in construction, civil engineering or structural inspection
What We Offer
- ✓Flexible part-time schedule, fully remote or hybrid from our East Sarajevo office
- ✓Clear path to a full-time position as the product grows
- ✓Own the computer vision core of a new product and shape its technical direction from day one
- ✓Work directly with the founder in a small, fast-moving team
- ✓Real-world impact: helping make building inspections safer, faster and more accurate
- ✓Competitive compensation based on experience
Ready to Join Us?
If this role sounds like a great fit for you, we'd love to hear from you. Apply now and take the next step in your career!