Resume

Search Intelligence (2024–2025)

Boosting Google Search with Language Models.

• •Fine-tuned a LLM for query understanding, improving the triggering of one Search feature by 20% in 7 languages.
• •Built an AI-based rater to evaluate Search results across multiple dimensions, speeding up the development cycle.
• •Developed a lightweight AI scorer to provide feedback signal for reinforcement learning of the main model.

Search Platforms (2022–2023)

Empowering the Google Knowledge Graph with Machine Learning.

• •Built and maintained large-scale data and ML pipelines using MapReduce, scraping and processing web data to feed the Knowledge Graph.
• •Served this knowledge data to 80+ Search clients, contributing to 25% of Google's global traffic.

• •Explored three different approaches to adapt transformer-based video inpainting models for real-time use.
• •Reached throughputs consistently above 20 fps with controlled loss in quality.

• •Collaborated with OECD to design a non-ML chatbot querying statistical data about countries.
• •Achieved above 90% accuracy on the set of queries typically asked by users of this data.

• •Developed mathematical models in Python to predict socio-political violence and instability around the world.
• •Designed a tool to estimate the number of French citizens abroad.