I build AI systems that plug into a company's real data, and I make them reliable: I know how to evaluate them, measure where they fail, and make them abstain rather than be wrong.

My edge is not "I can call an LLM." It's measurable reliability: retrieval evaluation, uncertainty quantification, calibrated abstention. My research background (conformal risk control) is exactly what makes this rigorous.

🌍 Neuchâtel, Switzerland · 🇨🇮🇨🇭🇫🇷 Ivorian, Swiss and French background 🎓 Dual Master's: Statistics (Neuchâtel) and ML/AI (Lyon 2)

A document-QA assistant for insurance contracts: the full chain, from a business question to a usable, traceable decision.

Not "a RAG that reads PDFs", but a system that knows when it misread, says so, reasons contract-by-contract to avoid mixing them, and measures its own reliability.

• Retrieval isn't the bottleneck: Recall@5 ≈ 0.97. The errors that remain have perfect retrieval, so the weak link is reasoning, not search.
• Contract confusion, measured: when several contracts share an attribute with different values, a global RAG mixes them. I isolated reasoning per contract and built trap questions to measure it: global 0/11 vs per-contract 11/11 correctly attributed, at about 2× the cost. A measured trade-off, not an assumption.
• Honest about extraction quality: each passage carries an explainable quality score (good / medium / poor) and why, and I document what it still misses (column interleaving, character spacing).
• Deployed, with a business-facing interface: decision, evidence, sources, uncertainty.

Stack: Python · sentence-transformers (E5) · Chroma · Groq · Streamlit · pdfplumber

• Reliable RAG & document AI: retrieval, grounding, citation, abstention
• Evaluation of AI systems: gold sets, the right metrics, and knowing when a metric lies
• Uncertainty quantification: conformal prediction (CRC, K-CRC, sem-CRC), calibrated abstention
• From prototype to deployment: turning a model into something a business can actually use

• Data Scientist intern, WTO, Economic Research and Statistics Division, Geneva (2026)
• Data Scientist (AI Research), Université Lyon 2 (2025)
• Statistical Research Analyst, LIVES Institute, UNIL (2022)
• Data Analyst & Product Designer, Assurland Africa (2021)
• Project Manager, ESN Neuchâtel (2023 to 2025)
• Mathematics Tutor, Anacours (2025) and Université Internationale de Grand Bassam (2019 to 2020)

Uncertainty quantification for medical AI: CRC, K-CRC and sem-CRC, with a benchmarking pipeline. The methodological backbone behind the reliability angle above.

Transfer-learning techniques for domain shift across data distributions.

Bias-mitigation strategies and fairness metrics on imbalanced datasets.

Languages: Python, R, SQL ML / DS: PyTorch, TensorFlow, scikit-learn, pandas LLM / RAG: sentence-transformers, Chroma, retrieval evaluation, answer-contract design Delivery: Streamlit, Git, deployment

M.Sc. ML/AI (MIASHS), Université Lyon 2 (2024 to 2025) Thesis: conformal risk control for semantic uncertainty quantification in CT imaging.

M.Sc. Statistics, Université de Neuchâtel (2021 to 2023) Thesis: statistical evaluation of public-prevention campaigns in Switzerland.

B.Sc. Mathematics, Université Internationale de Grand Bassam, Côte d'Ivoire (2017 to 2020) Thesis: decision trees: classification and regression cases.