AutumnOLYMPIAD
COLLECTION
RESEARCHED WITH AUTUMN
BA

Boammani Aser Lompo

ML researcher (MILA/ETS Montreal) - mechanistic interpretability, AI safety, learning theory; ENS/Agregation; IMO 2014 honourable mention
MILA / Ecole de technologie superieure (ETS), Montreal · Montreal, Canada · academic
IMO 2014 Honourable Mention (Burkina Faso)MSc ML Research At Mila/ETS (2025)Agregation (French National Maths Exam)ICML 2021 Co-author On Unimodal Ranking BanditsOpen-source AI-safety Projects (N2P, Guardrail)

Mathematics-to-AI researcher: after a competitive maths path in Burkina Faso (IMO 2014 honourable mention, PAMO 2015 gold) he studied at the French ENS and earned the Agregation before moving into ML research at Mila/ETS Montreal. His work spans mechanistic interpretability and AI-safety of model editing, reinforcement-learning theory (bandits, implicit bias), and clinical/multimodal NLP. As of 2026 he is openly seeking job opportunities and research collaborations.

Boammani Aser Lompo is a machine learning researcher at MILA / Ecole de technologie superieure (Montreal), working on the mechanistic interpretability of arithmetic circuits in LLMs (GPT-J, Llama-3-8B), the safety of model editing, the implicit bias of gradient descent under spurious correlations, and multimodal reasoning benchmarks. He holds an MSc in Machine Learning Research (MILA/ETS, 2025), an MSc in Mathematics, Vision & Learning (MVA, ENS Paris-Saclay, 2023) and an MSc in Fundamental Mathematics with the French Agregation (ENS Rennes, 2021), and previously interned at CREST/ENSAI and INRIA.

LinkedIn connections137
GitHub followers510 repos
Details
LocationMontreal, Canada
Company siteetsmtl.ca
UniversityEcole de technologie superieure (ETS) / Mila, Montreal
PythonC++JavaScriptBashPyTorchTransformerLensSAELensTransformersJAXvLLMLoRA/PEFTDockerWeights & BiasesMechanistic InterpretabilityAI SafetyReinforcement LearningLearning TheoryAI safety & mechanistic interpretabilityLearning theoryReinforcement learning / banditsFoundation modelsCompetitive chess (2000 Elo)Photography & content creation
Notes
  • Blog states he is 'actively seeking job opportunities' and open to research collaborations - the profile is current and he is on the market as of 2026.2026
  • Represented Burkina Faso at the IMO 2014 in Cape Town and earned an honourable mention, solving problem 4 fully (7/7, total 7, rank 446) - the result that put him on the international maths stage.Jul 2014
  • Current research pivots to AI safety and mechanistic interpretability (N2P project on GPT-J/Llama-3-8B circuits, direction-level exclusivity as an edit-safety criterion), aligning him with the interpretability wing of the field rather than pure NLP.2025
  • Transitioned from pure maths (ENS Rennes MSc + Agregation, ENS Paris-Saclay MVA) into applied ML at Mila/ETS Montreal, the standard French-mathematician-to-AI-research route and a signal of strong theoretical foundations.2023
  • Co-authored 'Parametric Graph for Unimodal Ranking Bandit' at ICML 2021 with Camille-Sovanneary Gauthier, Romaric Gaudel and Elisa Fromont during his CREST/ENSAI internship - an early publication tying bandit theory to ranking systems.2021
  • MSc research thesis 'Amelioration des modeles bases sur Transformers pour le traitement des nombres dans les documents medicaux' (ETS Montreal, defended May 2025) applied clinical NLP to numerical reasoning in medical records, supervised by Rita Noumeir and Philippe Jouvet.May 2025
  • Built and open-sourced applied AI-safety artifacts - a youth mental-health guardrail (9-signal distress taxonomy, F1=0.85) and Visual-TableQA - showing a pattern of shipping benchmarks/codebases alongside theory.2025
  • Holds the French Agregation (national competitive maths exam) - a credential that signals elite-level exam performance beyond the IMO and explains his ENS placements.2021
  • Named to Burkina Faso's IMO Hall of Fame and followed the 2014 honourable mention with a PAMO 2015 gold medal and a win at the 2015 National Mathematics Contest - a clean upward trajectory from olympiad competitor to the French ENS/Agregation pipeline.2015
  • Active on GitHub under Aser97 (10 public repos) and Hugging Face as AserLompo, with X/Twitter handle aserlompo - a builder-academic profile, not purely a paper-first researcher.2025
  • Funded by a 2024 R-SADC scholarship (~$23,000) and a 2018-22 French Government Excellence Grant (~$50,000), indicating institutional backing across his French and Canadian studies.2024

Competition record

Burkina Faso · IMO

2014 · Honourable mention · Rank 446