CV

Applied Mathematician

Paris · edimah.synesius-songo@proton.me · GitHub · LinkedIn · PDF

Bayesian inference, calibration, uncertainty quantification, sensitivity analysis and hypothesis testing, applied to model evaluations. Interest: measurements from inside the model that hold up under outside scrutiny.

Methods and tools. Python · R · SQL · C++ · Git, LaTeX · Bayesian inference & MCMC · calibration · uncertainty quantification · sensitivity analysis · permutation/bootstrap inference · multiple-comparison correction

Selected work

Detecting agentic collusion in public logs with statistical methods

Apart Research × CeSIA sprint, Sept 2026

Track 2. 14,591 reconstructed agent revisions and 32,656 live edits over four wikis, 2003–2026. A ladder of four null models, up to a difference-in-differences against an untouched sibling wiki with a placebo-in-time test: the incident weeks exceed all 4,872 reference weeks, the effect is 3.2 log points, the change point falls on 16 June. A weekly check built from the same features raises no false alarm over the 2,236 wiki-weeks of the preceding decade.

github.com/Edimah/agentic-collusion-detector-in-public-logs

A Calibrated Two-Level Detector for Secret Loyalties

Apart Research × Formation Research sprint, July 2026

With A. Randrianandrasana. My contributions: estimator, permutation test, calibration bench, difference-in-differences, control selection, activation probe. Calibration-matched controls cut false positives from 14/16 to 2/16 on a clean model, where every positive is false by construction. The instrument specification was written before any result existed and kept as a pre-registration, its result fields never back-filled. Reviewer: “This is rigorous work, and the calibration is a key contribution here.”

apartresearch.com/project/a-calibrated-twolevel-detector-for-secret-loyalties-g75x · github.com/Edimah/secret-loyalties

Modelling MNAR data with population data #1 : Why is this cell blank?

Blog post, July 2026

Suppression in published cancer-registry tables reads the count it hides, which makes the missingness not-at-random. Across tenths of expected case count the share of withheld cells runs from 93% to 0%; across tenths of population, from 13% to 1%.

edimah.github.io/blog/2026/why-is-this-cell-blank

Experience

Interim Data Lead

CRCDC Guadeloupe · Regional cancer screening · February 2026 – present

  • Rebuilding the regional cancer-screening data chain to deliver reliable data to stakeholders.
  • Delivered a data-chain audit and methodology documentation under regulatory constraints.

R, SQL, eSIS, DataQuery, Data Management, Health Data Regulation

Independent Consultant

Applied Statistics · AI Implementation · October 2025 – present

  • Feasibility analysis of AI tools in regulated health care (pharmacy) to facilitate professional workflows.

Data Management, Communication, Data Visualisation

Doctoral Researcher

IFP Energies nouvelles / ONERA / ENSAI · Île-de-France · November 2023 – September 2024

  • Multidisciplinary sensitivity analysis for complex physics systems.
  • Left the doctorate before completion.

R, Python, Sensitivity Analysis, Uncertainty Quantification

Research Intern

EDF R&D · Chatou · April 2023 – September 2023

  • M2 Thesis: Advanced methods for model calibration with uncertain inputs.
  • Designed adaptive, MCMC-based parameter re-adjustment strategies to handle uncertainty.

Python, OpenTURNS, Bayesian Inference, MCMC Algorithms

Education

  • 2023 — Master 2 — MASH (Mathematics, Machine Learning, Sciences, Humanities), Université PSL (Dauphine & ENS), with Merit.
  • 2022 — Master 1 — Applied Mathematics, Statistics track, Paris Dauphine-PSL.
  • 2021 — Bachelor’s — MEFA (Mathematics, Economics, Finance, Actuarial Science), Paris Dauphine-PSL.

Hobbies

String instruments · Bouldering · Muay Thai