R&D · AI / Photonics Engineer · Berlin
I build machine-learning systems for hard, real-world measurement problems: computer vision, predictive models, and the physics underneath. Currently working on ML-based non-destructive testing at BAM, Berlin.
I have a strong background in machine learning and photonics and have been active in science since 2016. My work sits at the crossroads of ML, data and applied physics: building predictive models, applying computer vision, and finding practical solutions to genuinely hard problems.
At BAM I lead work that uses ML and computer vision to detect internal defects in complex-shaped objects, connecting current research to real industrial use. Earlier I designed a concept for a miniaturised non-invasive glucose monitor — a lesson in innovation and resilience under tight resources.
I also use agentic AI workflows to shorten the path from hypothesis to validated result, automating data checks, feature exploration and reporting.
Method, experiments and signal post-processing to verify the internal integrity of waste containers without opening them — ML layered on top of the physics.
OutcomeML + classical post-processing raise defect detectability.
Optical polarimetry to measure blood-glucose without a needle: models from Mueller matrices, Monte-Carlo analysis, Python acquisition with lock-in and drift correction.
OutcomeA full medtech concept prototyped end to end.
Hands-on micro-optics assembly: coupling fibers, optical chips and resonators, plus scouting grant opportunities for the group.
OutcomePrecision optical alignment at chip scale.
LangChain and LangGraph compress hypothesis to validated result, with agents that automate data checks, feature exploration and reporting.
OutcomeA faster, more reproducible idea-to-insight loop.
Non-destructive integrity testing of nuclear-waste barrels: method & experiment design, excitation schemes, post-processing pipelines, and combining ML with conventional methods.
Micro-optics assembly — coupling fibers, optical chips and resonators — and research into grant opportunities.
Non-invasive polarimetry for blood-glucose: prototyping, models from Mueller matrices, Monte-Carlo analysis, Python automation, lock-in and drift correction.
Spectroscopy of carbon materials (XPS, Auger, EELS); a fast Langmuir-probe system; and modelling of magnetic confinement for a plasma thruster.
2024–2026 on ML-based non-destructive testing: thermography, defect detection, datasets.
Full list on BAM OPUS →An open thermographic dataset of nuclear-waste barrel cutouts with artificial internal defects — halogen long-pulse excitation, dual-camera thermal imaging. A reference for subsurface defect detection.
Averin, A., Hirsch, P. D., Lecompagnon, J.
BAM OPUS ↗Hatz, N., Averin, A., Lecompagnon, J.
BAM OPUS ↗Averin, A.
BAM OPUS ↗Averin, A.
BAM OPUS ↗Barretto, T., Gentes, S., Braun, J., Averin, A., Lecompagnon, J.
BAM OPUS ↗Open to ML, data-science and applied-research roles in Berlin or remote. If you have a hard data problem near the physical world, I'd like to hear about it.
© Anton Averin · AI / Photonics Researcher · Berlin