Alexandre Quemy

Building a multi-agent framework enabling software development in minutes. Lead AI engineer responsible for 90% of AI backend architecture and implementation in a 7-person startup. Participating in cloud infrastructure, MLOps, and product architecture alongside core AI development.

• Designed unified LLM client supporting 8+ providers (OpenAI, Anthropic, Gemini, Groq, Baseten, local via vLLM, ...), enabling vendor-agnostic AI integration
• Developed innovative function calling techniques beyond standard MCP with function dependencies, full OpenAPI specs support, and stateful tools for less tokens and API calls
• Built flexible declarative state-machine based agents with human-in-the-loop workflows
• Implemented robust one-shot learning and distillation "blueprint" mechanism for agent behavior replication
• Created knowledge extraction framework from user interactions and agent trial-and-error for continuous learning
• Improving SotA techniques for codebase representation with Knowledge Graph representation, unifying LSP, AST and additional contextual/metadata into a single graph (using Memgraph)
• Architected advanced RAG system with context engineering and rescoring techniques

Providing expertise in AI, notably LLM, and associated risks for European consortium dedicated to conversational AI. Assess technical proposals to ensure innovative value while mitigating ethical bias. Project funded by Horizon Europe, the EU's key funding programme for research and innovation.

Through my LLC, I accept some consulting missions related to AI, Data Science and Software Engineering.

• Clinical AI (InovIntell, 2024-present): Synthetic data generation for clinical studies with counter-factual analysis using Bayesian Networks, GANs, VAEs, and Neural ODEs
• Enterprise Search (Undisclosed, 2023-2024): Built semantic search engine and conversational SQL agent, designed SaaS architecture including cost estimation
• Financial AI (Undisclosed, 2023): Predictive modeling and recommendation system for M&A analysis using tripartite graph embeddings
• SEO AI (Watussi, 2017): Conceived and implemented NLP tool for Search Engine Optimization using Gensim and NLTK
• Successfully delivered 4+ AI projects with expertise in SBERT, LLM integration, Text-to-SQL, PostgreSQL, FastAPI, Docker, and Kubernetes

Led data science and AI team building SaaS platform for synthetic data generation and automated data quality assessment. Managed team of 6 engineers, shaped core IP, and drove product roadmap from research to production. Responsible for building the team, leading implementation effort, and answering fundamental research questions at fast pace.

• Co-author and maintainer of ydata-profiling library (10k+ GitHub stars), one of the leading Python tools for automated data profiling
• Researched and developed new generative models for synthetic data: scalable Normalizing Flow for tabular data, Bayesian network time series synthesis, SQL database generator with schema integrity, and longitudinal data augmentation (see Patent section)
• Built distributed processing systems using Dask, scaling models to process TB-scale datasets for enterprise clients
• Integrated LLM capabilities for table-to-text generation, enhancing synthetic data quality and interpretability

Led global financial crime detection models serving 200M+ users processing billions of transactions monthly. Managed complete model governance lifecycle from data quality to production monitoring.

• Designed and deployed main global Name Screening model used across HSBC group, reducing false positives by 70% while maintaining <2.0% false negative rate
• Automated data science reporting pipeline, achieving 90%+ time savings equivalent to 20 FTE/year in operational efficiency
• Developed controllable ML framework using rule injection, guaranteeing fixed model risk profiles for regulatory compliance
• Created multiobjective optimization methodology for model risk assessment, adopted as standard across compliance analytics
• Served as scientific advisor and peer reviewer while improving internal model governance lifecycle including scientific culture, code governance, and engineering risks

Architecture and engineering roles spanning cloud platforms, AI systems, and research collaborations. Responsible for flagship products and innovative AI platform development.

• Engineering on IBM Cloud Pak for Data flagship platform for Multi/Private/Hybrid Cloud serving 1000+ enterprise clients with a focus on monitoring, HA, predictive maintenance and D&R capabilities
• Built AI platforms for financial process automation (IBM Cobee) and code quality prediction (IBM Code Quality Center)
• Developed semantic search engine for legal documents and NLP/Topic Modeling services for enterprise applications
• Led monitoring solutions for GPFS and Red Hat OpenShift in hybrid cloud environments
• Served as Machine Learning and Data Science trainer, managed university collaborations, and supervised 10+ master students on AI research projects with conference presentations and lectures

Early career research collaborations across academic environments focusing on optimization and machine learning.

• Research Collaboration (Inria Saclay, 2013-2014): Evolutionary planner Divide-and-Evolve, MultiZenoTravel solver development, online parameter adaptation with Marc Schoenauer, TAO team
• Research Collaboration (Inria Nord-Europe, 2012-2013): Parallel Island Model for Metaheuristics, genetic algorithms, stochastic optimization with Clive Canape, DOLPHIN team
• Software Engineer Internship (Inria Nord-Europe, 2012): ParadisEO C++ framework development, shared memory parallelism module (SMP), continuous integration