Updated on 24 September 2026
Computer vision work package: defect detection, inspection stations, OCR and sorting, deployed on edge hardware
COO at Epoch8 LLC
Tbilisi, Georgia
About
We are an engineering team of 30+ machine learning and computer vision specialists. We build vision systems that run in factories, warehouses, and stores on the client's own hardware and under the client's own operating conditions. In a consortium, we take on one technical work package and carry it from the data protocol to the pilot.
WHAT WE DO
Visual quality control and defect detection. Surface defects, damage, deformation, contamination, missing or misplaced components, and packaging compliance. We are used to defect classes that are rare and expensive to collect, as is normal in industry.
Object and product identification. Barcode and label reading, OCR including embossed, engraved, and low-contrast text, and fine-grained classification across catalogues of thousands of visually similar items.
Inspection stations and gates. Multi-camera rigs that capture an object from several angles, reconcile the views into one decision, and write a structured record per unit.
Sorting vision. Product and material class recognition on moving conveyors for recycling and circular economy lines.
Retail shelf analytics. On shelf availability, planogram compliance, and product recognition at store scale, including fresh categories where the appearance of an item changes from day to day.
EDGE DEPLOYMENT
This stage is where most industrial vision projects succeed or stall, so we treat it as engineering rather than a final packaging step.
We compile and serve models on NVIDIA Jetson with TensorRT and DeepStream, and we optimize to a stated latency and power budget instead of to a benchmark score. That means we drive quantization, pruning, and architecture choices by the throughput the line actually needs, per camera, at the frame rate the process allows.
Running on the device has three consequences that matter in a European project. The system keeps working when the network does not. The cost per station stays predictable as the number of stations grows, because there is no per-frame cloud cost. And personal data can be reduced at the source, for example, by blurring faces before a frame is ever stored, which makes the data protection section of a proposal a description of the architecture instead of a promise.
WHAT WE CAN DELIVER IN A WORK PACKAGE
Data collection protocol and annotation guidelines, with the annotation carried out and quality controlled.
Dataset and benchmark design, including a public release where the project calls for one.
Model development, error analysis, and the evaluation methodology.
Edge integration and the serving stack on the partner's hardware.
Pilot deployment at a partner site and its operation through the pilot period.
Drift monitoring and scheduled retraining, built on Datapipe, our open-source framework for incremental data processing and retraining orchestration. It recomputes only what changed, which is what makes a pipeline affordable to rerun after the project ends.
The reusability and exploitation tasks that surround a vision work package.
WHERE WE HAVE APPLIED IT
Manufacturing and industrial inspection, contract logistics including automated inspection of pallets and cargo, retail and FMCG, recycling and circular economy, robotics and agriculture.
PRACTICAL
Our legal entity is registered in Georgia. Georgia has been associated with Horizon Europe since 2021, which means we participate and are funded on the same terms as entities from member states. Georgia is also named in the WIDERA work program among the associated countries with equivalent characteristics in terms of research and innovation performance, so our participation counts towards the widening and geographic balance of your consortium.
Working language: English. If you are assembling a consortium and the vision part is still open, send us the topic, and we will come back with a one-page work package outline.
Field
- AI-GenAI /Data/Robotics
Organisation
Similar opportunities
Expertise
- AI-GenAI /Data/Robotics
Olga Tatarinova
COO at Epoch8 LLC
Tbilisi, Georgia
Project cooperation
Reliable Human Feedback for Frontier Model Alignment
David Melhart
Assistant Professor at University of Southern Denmark
Odense, Denmark
Expertise
Advanced optical NDT and structural integrity for advanced manufacturing
- Photonics
- Manufacturing
- Other expertise
- Advanced Materials
- Testing & Analysis
Mauro Eduardo Benedet
Professor of Mechanical Engineering, Metrology & NDT at Federal University of Santa Catarina
Florianópolis, Brazil