I’m an AI / machine-learning engineer who builds ML systems that have to work in production: GPU inference services, evaluation infrastructure, data engines, and computer vision on edge hardware. I’m a physicist by training (MSci Physics, UCL; PhD, Imperial College London): 12+ years in engineering, 5+ in applied ML, 3+ shipping deep-learning systems that survive production, with a background in signal processing, embedded systems and distributed infrastructure.
I work as an independent contractor through my own consultancy, DevaLogic. Currently:
- R&D engineer at Hypodossier (Swiss document AI) — my primary engagement since 2023: GPU inference fleet, the internal ML labeling and evaluation platform, and the container-security programme.
- Software & AI lead on a counter-UAS computer-vision programme for a defense-tech client, alongside Hypodossier — ongoing R&D on detecting very small aerial targets in compressed video on NVIDIA Jetson-class hardware, where the evaluation harness and metric definitions were built before the detector.
- Founder of a drone-detection dataset engine — my own product and the training-data supply for the counter-UAS work above: a production ML data platform that turns publicly posted video into annotated training datasets (vision-language-model triage, embedding-based keyframe selection, segmentation-assisted annotation), running unattended.
What I care about
- Evaluation before optimisation. Golden datasets, detection metrics with proper confidence intervals, cost-per-inference analysis — measured claims, not demos.
- Systems that hold up in production. Distributed pipelines, reproducible GPU builds, security hardening, and the forensics (memory leaks, queue stalls) that come with running them.
- Working close to the physics. Signal processing, sensor and compression limits, small-target detection — problems where a physics background still matters.
Stack
PyTorch, CUDA/TensorRT, OpenCV, vision-language models (Gemini, Qwen2-VL, SigLIP 2), YOLO/RF-DETR, diffusion models, embeddings · Python, C/C++, Django/FastAPI · RabbitMQ, Hatchet, PostgreSQL/pgvector · Docker, Kubernetes/ArgoCD, Pulumi, AWS/GCP
Talks & community
I speak at Python and ML conferences — EuroPython 2025, PyCon Lithuania 2024/2025, Data Science Summit ML 2025 — and I founded and run the AI Code & Coffee Warsaw mentoring meetup.
The full CV is on the CV page (also as a PDF). Reach me at piotr.gryko@gmail.com or on LinkedIn.