Defect alerts with photo evidence
Operator workflow that turns camera footage into defect alerts, rejected-part photos, and a searchable inspection history.
CAMERADEFECTALERTHISTORY
Watch the demoDEMONSTRATION · OWN WORKSee the workflow
GSTREAM // COMPUTER VISION ENGINEERING
Custom computer vision for teams that have the footage, but not the CV engineer.
Detection, tracking and video pipelines built on your own data. Start with a fixed-price feasibility sprint. Real metrics before you commit to a build.
FIELD EVIDENCE
Selected Work
Selected work from the founder's prior roles.
Watch the demoAutomated Factory Inspection
Production vision system for automated defect detection and human-in-the-loop quality control.
97%Production precision
< 15msEdge inference latency
CAMERAINFERENCEQCREVIEW
PRODUCTION DEPLOYMENT - CUSTOMER CONFIDENTIALView Case Study
Multi-Camera Streaming & DeepStream Redesign
Redesign of a production video-AI stack for continuous multi-stream processing and GPU inference.
24/7Continuous operation
30 FPSLive pipeline
STREAMSDEEPSTREAMGPUML
CLIENT // ITARGET · STREAMING PLATFORMView Case Study
Dock timestamps from camera video
Prototype that logs every truck at a dock door: docked, loading started, loading finished. Runs on-prem.
CAMERADETECTTIMESTAMPBOOKING SYSTEM
PROTOTYPE · OWN WORKSee details
WHO IT'S FOR
Three kinds of teams
Image and video platforms
You sell software built on dashcam, drone, street-level or camera footage. You have data and developers. You need models that work on it.
Road and infrastructure inspection · drone inspection · property and insurance · agritech · video analytics
Operations teams adding vision
Your developers run your systems. You want cameras to scan, count or check something, and feed results into your ERP or app.
Warehouses · logistics · ERP-centric operations
Manufacturers with variable defects
Your camera inspection misses defects or rejects good parts, because no two defects look the same. Rules can't separate them.
Painted and coated parts · wood · print · molded plastic and glass · textiles
PACKAGES
Fixed scope. Clear output.
Every project starts small. You see results on your own data before any larger build is quoted.
01START HERE
Feasibility Sprint
A baseline model trained on your sample images or video.
- Real metrics on your data
- Where it fails, and why
- Go / no-go decision
- Build plan and fixed quote
€2,500–4,000FIXED · 1–2 WEEKS
02OR START HERE
Model Audit
Your model or vision system underperforms. We find out why.
- Data, model or hardware: which one
- False rejects and misses, measured
- Fix, or a concrete plan
€2,500–4,000FIXED · 1–2 WEEKS
03
Build
Production models and the pipeline around them.
- Video → detection → tracking → events
- API or database integration
- Edge or cloud deployment
- Training code, docs, handover
€10–40kPER SCOPE · 4–10 WEEKS
04
Care Plan
Models drift when cameras, products or conditions change.
- Performance monitoring
- Retraining on new data
- Fixes when conditions change
€1–3k / monthONGOING
Not a fit: simple rule tasks like label or part presence. Your Cognex or Keyence integrator does these well.
01 SEND SAMPLES02 SPRINT OR AUDIT03 BUILD04 CARE
You only commit to the next step.
Technical Founder's Note
One engineer, from data to deployment.
“I lead every engagement personally. You talk to the person who builds your system.”
8+ YEARS APPLIED MLCOMPUTER VISIONVIDEO ANALYTICSEDGE INFERENCE
Kaspars Polis
LinkedIn ↗Have footage and a problem? Send 20 images.
Within 48 hours you'll hear whether it's detectable, and what a sprint would cover.
Send images