Skip to content
View ysmouhib's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report ysmouhib

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ysmouhib/README.md

Younes Mouhib

MSc Mathematics, ETH Zürich — graph theory, combinatorics, optimization, probability, verified computation. Record lower bounds for Hales–Jewett numbers (HJ(3,3) ≥ 22, HJ(4,2) ≥ 14), sole-author preprint arXiv:2606.22155, every certificate machine-checkable.

Looking for an entry-level role or internship in data science, optimisation or quantitative analysis: banking, fintech, analytics, operations research, risk, or Python/data positions. Open to relocate, immediately available.

I build with AI coding agents and keep them honest with tests, pre-registered evaluations and formal checkers.


Selected work

  • ⚽ Football injury risk — weekly model · live match-day tool. Weekly injury-risk scores for every player in Europe's top five leagues, from public data only: pre-registered and tested once on a season it had never seen, the top-decile risk group was injured 6× as often as the bottom decile. Plus an offline bench app that tracks each player against his own normal during a match, built and validated end to end on synthetic matches (no club data yet).
  • 📉 near-momentum-bot — Cost-aware backtesting and walk-forward validation for Binance spot. The honest answer included: when the edge doesn't survive fees, the report says so. Browser simulator →
  • 🧩 sudoku-vision — Reads handwritten sudoku with a small CNN, solves with SAT, and repairs misread digits with MaxSAT: 75% → 99% of grids read perfectly.
  • 🤖 brigade — Multi-agent LLM math prover: agents propose, only an independent Lean 4 + Mathlib verifier accepts. No proof, no merge. Project page →
  • 🪳 cockroach-smash — A live multiplayer browser game used as a testbed for invisible-prompt defenses: hidden Unicode instructions (ASCII smuggling, emoji smuggling, Trojan Source) are stripped in Postgres before anyone, human or AI, reads them. 20 attack payloads in CI; the tests found and closed 4 gaps; live site 29/29. Play →
  • 🔒 hj-certificates — SAT encoders and machine-checkable certificates for record Hales–Jewett lower bounds. Stdlib-only verifiers, DRAT-validated flagship refutation, CI on every push.
  • 🌐 hales-jewett-explorer — Interactive 2D/3D companion to the thesis: paint colourings of [t]ⁿ and watch the combinatorics update live; every counter re-verified in the browser. Live demo →

Find me

Pinned Loading

  1. football-injury-risk football-injury-risk Public

    Weekly injury-risk scores for every player in Europe's top five leagues from public data. Calibrated logistic regression, one-shot holdout: top 10% injured 6x as often as bottom 10%.

    Python

  2. Football-live-injury-risk Football-live-injury-risk Public

    Live, player-relative injury risk for football bench staff: an offline match-day app with a calibrated hazard model and budgeted alert cards, validated on synthetic seasons. Weekly half: football-i…

    Python

  3. near-momentum-bot near-momentum-bot Public

    Cost-aware backtesting & walk-forward validation for Binance spot — a study in honest evaluation, not a performance claim

    Python

  4. sudoku-vision sudoku-vision Public

    Reads sudoku images with a neural network and solves them with SAT; MaxSAT repairs misreadings by finding the most probable reading with exactly one solution.

    Python

  5. brigade brigade Public

    Multi-agent math prover — agents propose, Lean 4 disposes

    Python 1

  6. hj-certificates hj-certificates Public

    Machine-checkable certificates and SAT encoders for record Hales–Jewett lower bounds (HJ(3,3) ≥ 22, HJ(4,2) ≥ 14). Stdlib-only verifiers re-check every claim in minutes. Companion to arXiv:2606.22155.

    Python