Léa Lacoste

– Build AI Assurance assistant backend from scratch, including agent to recuperate insurance documents from insurance portals, ingestion of assurance policy documents, chatbot to answer questions about insurance coverage. – Tech stack: Langgraph, Instructure, AWS Bedrock, Python, Pydantic, Postgres, FastAPI

– Developed LangChain-based chat agents. – Optimized search relevance and performance on OpenSearch. – Built a robust data pipeline integrating web scraping and LLM based data extraction. – Contributed to the design and deployment of scalable, production-grade AI solutions. – Tech stack: Python, LangChain, LangGraph, LangSmith, OpenSearch, GCP, DataBricks.

– Owned and optimized the core search infrastructure (Elasticsearch on AWS), reducing compute costs by 90% and memory usage by 75% while improving relevancy. – Implemented architectural changes to improve performance and reduce AWS costs, including optimized data retention in S3 and Postgres and improved event handling. – Participated in integrating OpenAI APIs, combining retrieval from web and Elasticsearch with LLM-based processing to deliver insights on large company datasets. – Contributed to the launch of Tactic Generate, a Retrieval-Augmented Generation (RAG) product enabling users to upload sets of documents, receive summaries, and query content via natural language.

– Configured and optimized search services for Rakuten's global eCommerce platforms. – Built and maintained a PySpark-based KPI tool analyzing massive datasets across 20+ services and three data centers (JP, EU, US), supporting A/B testing and strategic decision-making. – Led integration of Fits.me technology into Rakuten's fashion sites, solving a critical speed bottleneck and securing the project's success — contributing to a R&D tax credit. – Supervised interns and collaborated with international, cross-cultural teams.

– Developed a semantic search engine based on Solr, using linguist-defined semantic rules compiled into Finite State Transducers (FSTs). – Improved FST efficiency by identifying and collapsing redundant paths, and built a knowledge graph to reveal connections between named entities. – Owned build and release processes with Maven and maintained detailed technical documentation to support long-term maintainability. – Tech stack: Solr, Lucene, Java, Maven, Linux

– Developed a 3D video analysis extension of Definiens' 2D image analysis software for medical and biological research. – Collaborated with Dr. Carl-Phillip Heisenberg (MPI) and Dr. Gerd Binnig (Definiens) to design rule sets for detecting cell growth in 3D microscopy videos.

– Developed a 3D video analysis extension of Definiens' 2D image analysis software for medical and biological research. – Collaborated with Dr. Carl-Phillip Heisenberg (MPI) and Dr. Gerd Binnig (Definiens) to design rule sets for detecting cell growth in 3D microscopy videos.

– Designed countermeasure algorithms to detect spoofing attempts in iris biometric systems. – Acted as the primary technical liaison for clients and partners.

– I was in the group PRIMA while working on my PhD.