Computer vision that is graded inside the product
I am an AI engineer and consultant. I build computer vision systems: object detection on drawings and documents, OCR, and near-duplicate image identification. I grade each model inside the product, not only on the training metric.
I am based in Dallas-Fort Worth. I work with clients in Dallas-Fort Worth and Oklahoma City, remotely or on site.
I reply within two working days. NDA on request.
Ways to start
Each offer has a fixed scope and a clear result. You get a quote after the intro call.
-
Computer Vision / AI Agent Prototype
3-4 weeks- For
- Teams that want to test an AI idea on their own data.
- You get
- A working demo with metrics and a clear go / no-go note.
-
AI Feature Audit
3 days- For
- Teams with an AI feature that feels unreliable.
- You get
- A written review of the feature, its data and its tests, and a fix plan in priority order.
What I have built
Client work uses codenames. I can share more detail under an NDA.
- A detection pipeline (YOLO, then RF-DETR) that finds HVAC symbols on construction drawings. The best detector reached mAP50 0.755.
- A correction flywheel: every human fix becomes new training data.
- Vector-text reading and tiled OCR to count grille tags from the PDF itself.
- Revision tools: slip-sheeting, FFT phase-correlation auto-align, added/removed diff.
- As a core engineer on a small team: the identity layer that finds the exact printing of a collectible in a 151,260-item catalog.
- Model evals that score precision, recall, F1 and count error per class, on full pages.
See the proof: Blueprint Vision: Computer vision for construction takeoff and HVAC estimatingCollectible ID: Near-duplicate image identification and pricing at catalog scale
Questions
- What computer vision work have you done?
- I trained YOLO and RF-DETR detection models that find HVAC symbols on construction drawings for takeoff and estimating. My best detector reached mAP50 0.755. On a small team, I also built near-duplicate identification across a 151,260-item catalog.
- How do you measure a vision model?
- I grade the model inside the product, on full pages, per class: precision, recall, F1 and count error. The training metric says that training converged. It does not say what a user gets.
- What limits the accuracy of a detection model?
- In my work, data quality limited accuracy more than architecture. So I change the product, so that each human correction becomes clean training data.
- What does a computer vision prototype include?
- A working demo on your own images or documents, with metrics and a clear go / no-go note. It takes three to four weeks.
- Where do you work?
- I am based in the Dallas-Fort Worth area. I work with clients in Dallas-Fort Worth and Oklahoma City. I also work remotely with teams anywhere, and I travel to work on site when it helps.
Tell me what you are building.
30 minutes. Tell me what you are building. I reply within two working days. Client details can stay private under a mutual NDA.