Steven Bucaille
Senior Machine Learning Engineer - Computer Vision & Applied AI
SB
About
Machine Learning Engineer with 4+ years of experience building and deploying computer-vision systems for industrial applications. Delivered 12x inference speedups, 99.9% model accuracy, and ML workflows managing 33,000+ images and 150,000+ annotations. Active Hugging Face contributor with experience across PyTorch, MLOps, production AI delivery, and applied Generative AI.
Work Experience
Sep 2025 - Present
Machine Learning Engineer
Technical owner on client AI projects, from discovery and data collection through model development, application delivery, and production deployment.
- Delivered production systems for industrial inspection, wind-turbine monitoring, and bilingual voice automation
- Rolled out Ray internally for six engineers, with DVC/S3 workflows
- Shared implementation practices and project lessons through internal knowledge-sharing sessions
Projects
- Industrial packaging seal inspection: Deployed a seal-inspection application running 24/7 on one production line, with around 200 uses per month. Built an RGB-camera capture station with controlled lighting and 3D-printed components; collected about 500 physical samples. Trained classification, segmentation, and detection models with Lightning and MLflow; delivered the software and automatic startup, including on-site redeployments.
- Multi-camera live image stitching for hockey broadcasting: Built live multi-camera hockey stitching with automated keypoint configuration.
- Wind-turbine status extraction: Extracted status from four dashboards in parallel every 30 seconds using OCR and color processing, writing results to PostgreSQL. Deployed in May 2026 and still in production.
- Bilingual real-time voice agent: Delivered a production voice agent using Twilio, OpenAI Realtime, Azure, and Docker, with API/PDF retrieval and human transfer. Reduced calls handled per agent from 80 to 12 per day; the system handled about 1,200 calls in one week.
Python
PyTorch
Lightning
MLflow
Ray
PostgreSQL
OpenAI Realtime
Azure
Docker
Hugging Face Transformers & Datasets
Computer Vision
DVC
Gradio
3D Printing
Client Communication
Discovery & Requirements Gathering
Cross-functional Collaboration
Iterative Problem Solving
Ownership
2022 - 2025
Machine Learning Engineer
Machine Learning Engineer in a research team of 10 people. Owned the architecture and model work for embedded industrial computer-vision products, while coordinating data acquisition, labeling, and reproducible delivery workflows.
- Implemented GitLab CI, DVC, Ray, and CML for reproducible training, experiment tracking, and model delivery.
- Coordinated data acquisition and labeling for more than 33,000 images and 150,000 annotations within a team of 15 people.
Projects
- Camera-shift compensation for AGV positioning: Recentered fixed image-analysis zones when autonomous vehicles presented parts with camera misalignment, reaching 99.9% accuracy. Reduced local inference from 250 ms to 20 ms with TensorRT, quantization, batching, and cached reference features.
- Real-time multi-camera vehicle detection and tracking: Rebuilt the production pipeline around RT-DETR, tracking a reference vehicle across up to four cameras at 30 FPS. Unified fragmented libraries into a reusable production library with batched inference and direct product integration.
Python
MLOps
Lightning
Hydra
Ray
ONNX/TensorRT
Gradio
AWS
GitLab CI
Cross-functional Collaboration
Problem Solving & Optimization
Initiative & Ownership
Business Understanding
Communication
April 2019 - July 2019
Research Intern
Researched non-functional requirements in published APIs using a distributed primary-secondary architecture and Google Cloud Platform.
- Researched "Discovering Non-Functional Requirements from Published APIs" under the supervision of Dr. Jordi Cabot
- Analyzed API properties from different locations using a distributed primary-secondary architecture and Google Cloud Platform
- Published the work at ICWE 2020, receiving the Best Demo & Paper Award
Angular
NodeJS
C3.js
Google Cloud Platform
Open Source Contributions
Implemented SuperPoint, SuperGlue, LightGlue, EfficientLoFTR, and RF-DETR, a state-of-the-art object-detection model, into the Hugging Face Transformers library, expanding its computer-vision capabilities. These contributions have reached over 32M total downloads.
Projects
The agent capable of identifying and using the best model among the entire Hugging Face Hub for your computer-vision needs.
smolagents
Modal
Claude
RAG
Education
Katholieke Universiteit Leuven
2019 - 2023
Master's Degree in Computer Science : Major Artificial Intelligence
Thesis : Data-driven optimisation of substitutions in football
Université de Lille
2018 - 2019
Bachelor's Degree in Computer Science
References
“I am writing to recommend Steven Bucaille for a position that requires strong technical skills, a deep understanding of computer vision, and a collaborative approach. [...] Steven's contributions are technically sound, well-documented, and user-friendly.”
View the recommendation letter
“His solid knowledge in web technologies and strong programming skills allowed him to design a non-trivial solution that fulfill initial requirements of this project. [...] In summary, I'm confident in stating that he has a strong potential and consider him a valuable candidate for your organization.”
View the recommendation letter
“... once again, Steven went beyond the objectives I set, he demonstrated a quality that I did not detected before: the ability to deal with the different stakeholders of a project. [...] He then created a surprising and very valuable group dynamics.”
View the recommendation letter