ML Engineer - video & industrial data

About the company

PackGuru is an AI platform for production that helps FMCG and cosmetic plants better understand what is happening on their packaging lines and react faster to problems.

Machine knowledge often lives in the heads of a few experienced people. PackGuru turns raw signals and scattered knowledge into concrete instructions, available to everyone on a shift, including in plants where local experts are lacking.

We work closely with Unilogo, a manufacturer of packaging lines and machines. Thanks to this, our solutions are created with full access to knowledge about machine design and their control software, rather than solely from the "outside." Our system operates in production in factories of global FMCG corporations in the United States, Mexico, and Brazil. These are real lines, real data, and a real impact on production results.

We are a small, dynamic team that is growing very rapidly.

With us, there are no long decision chains. Your work goes into production within weeks, not quarters, and you have a real impact on the direction of product development.

We use LLM intensively, both in the product itself and in the team's daily work, but we do it sensibly: where they actually provide an advantage, not just because it is trendy.

Where classical ML, deterministic logic, or a well-designed algorithm works better, we choose those instead.

Your tasks

We combine machine data (PLC), images from cameras analyzed on NVIDIA edge devices, and technical plant knowledge (documentation, procedures, resolved failure cases) into a single system that supports operators, technicians, and engineers in their daily work.

Why do we do this? Factories today have plenty of data, but they still lack answers to simple questions: why did the line stop, what was the cause of the loss, and how to fix it before it repeats.

• Maintaining and developing data pipelines from production lines: PLC data, camera streams, and video processing on NVIDIA devices.
• Designing, training, and deploying video analysis algorithms running on edge devices (Jetson), including their optimization for performance and reliability.
• Building a pipeline for subsequent ML models on machine data, e.g., parameter trend analysis and drive condition assessment based on torque, from data exploration to production deployment.
• Maintaining and developing the LLM and knowledge graph layer, including regular quality evaluation of the AI assistant's responses.
• Co-creating the telemetry processor: interpreting machine signals, defining machine states and events, detecting causes of losses on the line.
• Monitoring model performance in production, detecting degradation, and planning retraining.
• Collaborating with the automation team and client-side engineers on data collection and result validation.
• Participating in deployments on clients' production lines, both remotely and on-site.
• Documenting solutions and sharing knowledge within the team.

Requirements

We are looking for someone who wants to build systems that operate in the physical world, from the camera and machine controller to the model and AI assistant, and who is not intimidated by the combination of software engineering and the production floor.

_Must-have:_

• Min. 3 years of experience as an ML Engineer, Computer Vision Engineer, or in a similar role, with models running in production, not just in a notebook.
• Proficiency in Python and practical experience with PyTorch.
• Experience in computer vision: detection, segmentation, or video analysis (e.g., YOLO or similar architectures).
• Experience with NVIDIA Edge platforms (Jetson) and model optimization for inference (TensorRT, ONNX, quantization).
• Knowledge of video stream processing (GStreamer, DeepStream, RTSP, H.264/H.265 encoding).
• Experience in building and maintaining data pipelines, including time series data (PostgreSQL / TimescaleDB or similar).
• Comfortable working in Linux, with Docker and Git.
• Independence and the ability to see tasks through to completion in a small team.
• English language at a level that allows working with documentation and communicating with foreign clients.

_Nice to have:_

1. Experience with industrial data: PLC, OPC-UA, MQTT, machine signals.
2. Knowledge of time series analysis methods, anomaly detection, or predictive maintenance (e.g., parameter trends, drive condition).
3. Experience with LLM, RAG, or knowledge graphs, including quality evaluation of AI assistant responses.
4. Knowledge of C# / .NET.
5. Experience with cloud (GCP) and the MLOps approach: model versioning, monitoring, automatic retraining.
6. Experience working on production lines or in a factory environment (FMCG, packaging).
7. Readiness for occasional trips for client deployments, including abroad.

Must have: Machine learning, Git, Video processing, Python

Nice to have: Docker, GCP, OPC-UA, MQTT, C#