Steven Bucaille
Research Engineer in Artificial Intelligence
SB
About
Enthusiastic and results-oriented Research Engineer with 2 years of experience leveraging computer vision for industrial quality control tasks. Proven ability to design, develop, and deploy robust vision systems, achieving significant efficiency gains and accuracy improvements. Possesses strong problem-solving, data analysis, self-learning and communication skills, adept at collaborating with cross-functional teams.
Work Experience
2022 - 2024
Research Engineer
Currently working on self-supervised learning using AWS computing power
- Deployed a scalable an auto-configuring Ray Cluster on AWS overcoming the limits of on-premise single GPU
- Used a vector database to extract the most diverse dataset among 1.3M images
Product feature to improve robustness to object position variability
- Analyzed feasibility with image matching PoC.
- Optimization of the deep learning model to comply with timing and embedded hardware requirements : Pruned unused layers, maximized GPU usage, minimized GPU ⥦ CPU/RAM data transfers, Optimized inference using TensorRT, achieving 12x speedup (from 250ms to 20ms)
- Benchmarking on industrial data and push to production
- With a 99.9% accuracy, the feature delivers near-perfect results, with less than 1 error in 1000 cases.
- Developed a Gradio dashboard for explainability of the final algorithm
Led data acquisition and labelling across projects
- Acquired on-site data and launched a company-wide initiative to label 1400 images (text/QR codes). Non-technical staff successfully labeled the data in under 2 weeks, securing the company's competitive edge
- Established a year-long partnership with a third-party labelling company for faster development.
- Oversaw labelling of 33,000 images (150,000 annotations) across various formats (image, video, text, classification).
Package content compliance automation use-case
- Recorded 2.5 hours of on-site video with 3 cameras
- Labeled first 15 minutes using CVAT to infer package count and content (1100 packages)
- Trained classifier to identify packages requiring human action, filtering 46% automatically using image segmentation
Python
Lightning
Hydra
Ray
ONNX / TensorRT
Gradio
AWS
April 2019 - July 2019
Research Intern
Conducted research on "Discovering Non-Functional Requirements from Published APIs" under Dr. Jordi Cabot supervision
- Designed a web app using a Master-Slave architecture
- Analyzed non-functional properties (latency, uptime) of published APIs across geographically distributedlocations (Europe, America, Asia, ...) using Google Cloud Platform
- Developed a user-friendly dashboard to display analysis results
- Work culminated in a paper published at the ICWE 2020 conference, which received the Best Demo & Paper Award.
Angular
NodeJS
C3.js
Education
KU Leuven
2019 - 2023
Master's Degree in Computer Science : Major Artificial Intelligence
Thesis : Data-driven optimisation of substitutions in football
Computer Vision
Natural Language Processing
Speech Recognition
Data Mining
Principles of Machine Learning
Design of Software Systems
Declarative Languages
Distributed Systems
Software Architecture
Modelling of Complex Systems
Fundamentals for Computer Science
Comparative Programming Languages
Fundamentals for Artificial Intelligence
University of Lille
2018 - 2019
Bachelor's Degree in Computer Science
Projects and contributions
HuggingFace Transformers
github.comhuggingface/transformers/pull/28966/
Implementation of the SuperPoint model to the library
Contribution
Python
Lightning GlueFactory
github.comsbucaille/lightning-glue-factory
Pytorch Lightning implementation of GlueFactory
Side project
Python
Lightning
Hydra
Side project
Candle
Rust