🦾
Sr. AI Engineer @ IBM
AI/ML Engineering • DataOps & MLOps
GenAI (Text, Image, Audio) • Deep Learning • Machine Learning
I'm a Senior AI Engineer at IBM, passionate about crafting robust, large scale, and high-impact AI systems.
With 4+ years of experience, I specialize in transforming cutting-edge research into production-ready solutions
that deliver measurable business value.
while True:
raw_data = ingest_from_sensors()
data = clean_and_transform(raw_data)
insights = run_ml_pipeline(data)
deploy_action(insights)
[PSEUDO CODE]
Synergy of bachelor's in Microengineering and Master's in Data Science & Robotics
Scaled fine-tuning & agentic framework into a global product used across IBM as a Core Dev & Tech Lead.
Optimized GPU finetuning (SFT, DPO, GRPO, LoRaX) and built async multi-threading agents with RAG and MCP integration using Kubernetes, vLLM, and DeepSpeed.
Co-led IBM Synthetic Data generation and delivered 20+ enterprise use cases — eg. agentic CI/CD, Neo4j-based KYC, and SQL-generating — driving multi-million-dollar impact.
RAG over 1M+ legal docs, featured in Microsoft Customer Story
Optimized terabyte-scale ETL and ML pipelines, reducing compute costs by 30% for hedge-fund strategies.
Fine-tuned FinBERT for NER and text classification.
Deputy Tech Lead (MLOps), guiding 10+ developers.
Research project at EPFL's LASA Laboratory (Prof. Aude Billard) on real-time robotic manipulation using auditory and tactile inputs. Developed an LSTM–CNN model to enhance grasp stability in high-speed industrial robots, in C++ on ROS.
Led a deep learning project for early tremor detection, achieving an F1-score of 0.91 using LSTM–CNN models. The system analyzed keyboard and mouse dynamics to predict neurological disorders, with focus on SHAP interpretability.
Designed MLOps pipelines, A/B testing, and optimization.
Contributed for two years to developing a sports connectivity platform at a dynamic startup, analyzing user data to drive 50,000+ user growth and secure funding in Lausanne.
Teaching in Maths, Physics, and Biology in Primary School at College Arnold Reymond.
Master's Thesis at ETH Zurich (Prof. Roger Wattenhofer) — developed a robust multimodal contrastive framework for vision-language transformers, enabling image–text retrieval and enhancing cross-modal alignment through adversarial self-supervision. Trained large-scale models (50 x Nvidia V100) using PyTorch, DeepSpeed, SLURM.
Compared two policy-gradient methods, Actor-Critic and Advantage Actor-Critic (A2C), approaches for Pong training in PyGame.
I believe the next evolution of Artificial Intelligence lies not in larger models, but in integrated intelligence — efficient systems that run at the edge, embedded across our devices.
LLMs and MLLMs are becoming the computational kernel of modern operating systems, orchestrating data, tools, and actions in real time while adapting to each user’s context and preferences.
By compressing and distilling knowledge, we bring reasoning closer to where information lives — enabling generative and traditional ML to work seamlessly together.
This shift will redefine our relationship with technology, transforming devices into intelligent collaborators that act, learn, and evolve with us.
Proud to help build this new era of intelligence. 🚀
