Abdallah Hakim Kone
Machine Learning Engineer at BNP Paribas
I take machine learning models to production and keep them running.
MLOps and infrastructure
- Deployment pipelines
- Networking
- Model monitoring
- Docker
Modeling
- LLMs
- RAG
- Speech-to-text
- NLP (BERT)
Production
About
At BNP Paribas, data scientists build the models; I build everything that takes them to production: deployment pipelines, serving, monitoring, and the infrastructure and network work around them. I have worked in system and network administration, software development and machine learning. That is why I am often the link between teams: infrastructure, data scientists, developers. Before BNP, I spent four and a half years at Softeam on production computer vision and tabular models for a public-sector client.
Outside work, I love history, art and heritage. I like to visit historic sites and look at what past civilizations left behind. I photograph these places. I also brew specialty coffee.
Projects
Personal projects built end to end, from the problem to the results, limits included. My production work is confidential; it is described under Experience.

Content-based image retrieval
Best MAP 0.67 on colour and style, with SigLIP 2 at 25 ms per query
2023 · PyTorch, timm, OpenCV, scikit-learn, Polars
A benchmark of eight feature extractors and three distance measures for finding similar clothing images, measured with MAP and MRR.

Stochastic MuZero for 2048
Reached tile 2048 in 2 of 16 games after 6.8 h on one GPU (about $4); mean score 13,446 against 2,616 untrained
2026 · JAX, Flax, Optax, mctx, NumPy, RunPod
A reinforcement learning agent for 2048 that plans with a learned model and tree search over random tile spawns, written in JAX.

Maze-solving agents on MiniGrid
In progress · Two agents and a four-level maze curriculum built; training runs not recorded yet
2025 · TensorFlow, Keras, TensorFlow Probability, MiniGrid, Gymnasium, Pydantic
PPO with curiosity-driven exploration and Rainbow DQN, trained from pixels to navigate multi-room mazes of increasing difficulty.

Deep learning image watermarking
40.1 dB PSNR and 1.2% BER without attack, against 40.6 dB and 0.7% in the paper; rotation not learned
2026 · PyTorch, Kornia, uv
A PyTorch reproduction of a CNN watermarking paper: two networks trained together against the paper’s attack mix, measured table by table against the published results.
Experience
since 06/2025
Machine Learning Engineer, BNP Paribas
I take a data science team’s models to production. We host every model in-house for security reasons.
Nantes, France
- Package the team’s models as APIs and batch jobs in Docker images, deployed on Domino Data Lab with GitLab CI/CD.
- Serve open-weight LLMs (Llama, GPT-OSS, Gemma, Mistral Medium), Whisper and Voxtral for speech-to-text, and the data scientists’ BERT models.
- Build pipelines that transcribe conversations, then write a summary or check control points.
- Build RAG systems with BGE-M3 that suggest replies to customer emails, first on ChromaDB, now on an embedding store I wrote.
- Handle the network side of deployments and keep the model APIs reachable.
- Set up model monitoring and alerting with Arize and Kibana.
Python, Docker, GitLab CI/CD, Domino Data Lab, Transformers, ChromaDB, DVC, Arize, Kibana
01/2021 – 06/2025
AI Engineer (consultant), Softeam
Consultant for a major public-sector client, automating the processing of traffic violations from radar images with computer vision and tabular models.
Rennes, France
- Rebuilt the vehicle identification model as a mixture of specialized experts, which reduced recognition errors.
- Designed a Swin Transformer classifier that rejects unusable radar images before processing.
- Developed heuristics that locate the violation zone in radar images.
- Automated retraining, from data collection to model selection (Optuna, Aim, MLflow).
- Improved the sampling of images sent for labeling, and added bootstrapped confidence intervals to the business metrics.
- Integrated the models into the production processing chain with the teams that run it (requirements, acceptance tests).
- Designed a secure training environment with architects and the CISO, and led GDPR compliance with legal experts.
- Mentored three junior engineers.
Python, TensorFlow, XGBoost, SHAP, OpenCV, Optuna, MLflow, Aim, FastAPI, Docker, RabbitMQ, MongoDB, GitLab
09/2019 – 09/2020
Data Scientist (work-study), Orange Business Services
Built computer vision and simulation tools to analyze retail performance for a shopping-center operator.
Rennes, France
- Benchmarked multi-object tracking models (Deep SORT, YOLO, SSD) to produce in-store retail KPIs.
- Built and deployed a similar-face search application.
- Prototyped a multi-agent reinforcement learning simulation of customer behavior.
Python, TensorFlow, PyTorch, OpenCV, Flask
04/2018 – 08/2018
AI Developer (internship), Compagnie Ivoirienne d’Électricité
Abidjan, Côte d’Ivoire
- Benchmarked object detection models (YOLO, SSD, Faster R-CNN) to detect incidents on the power grid.
- Built a cross-platform mobile app to manage grid incidents.
- Built an app that reads electricity meter indexes.
Python, TensorFlow
04/2017 – 09/2017
IT intern, Veone
Abidjan, Côte d’Ivoire
- Deployed monitoring systems and managed services (Centreon).
Linux, Bash, Python, Centreon
Skills
- Modeling
- LLMs, RAG, Speech-to-text, NLP (BERT), Computer vision, Tabular models, Reinforcement learning, Transformers, TensorFlow, PyTorch, scikit-learn, XGBoost
- MLOps and infrastructure
- Deployment pipelines, Networking, Model monitoring, Docker, Linux, GitLab CI, FastAPI, MLflow, DVC, Domino Data Lab, Mistral AI Studio, Arize, Kibana
- Languages and data
- Python, Rust, TypeScript, pandas, Polars, NumPy, MongoDB, Elasticsearch, RabbitMQ
- Spoken languages
- French (native), English (professional)
Education · Certifications
2018–2020
General engineering degree (Diplôme d’ingénieur)
IMT Atlantique, Brest, France
Mathematics and signal processing, data science, software engineering, game theory.
2014–2017
Bachelor’s degree in computer networks and systems
ESATIC, Abidjan, Côte d’Ivoire
Networks and network administration, linear algebra, signal processing, statistics.
2025
MLOps | Machine Learning Operations
Duke University
2023
Machine Learning Engineering for Production (MLOps)
DeepLearning.AI
2021
Deep Learning Specialization
DeepLearning.AI
2020
Advanced Machine Learning with TensorFlow on Google Cloud Platform
DeepLearning.AI
2020
TensorFlow Developer Certificate
TensorFlow Certificate Program
2020
L’Atelier RGPD
CNIL
2019
Machine Learning
Stanford University