Tommy Lepesteur

I design artificial intelligence pipelines applied to drug discovery. My field: the frontier between classical computational chemistry (docking, ADMET, ChEMBL) and recent protein deep learning methods (ESM-2, ESM-IF1, ProteinMPNN, Rosetta ddg).

> At Delta G Double Dagger Therapeutics, I generated and prioritized over 6,500 candidate molecules across several therapeutic families. The most mature projects are protected by patents filed with INPI.

> I publish my methodology in open science via Zenodo (8 preprints with DOI), because reproducibility is the only viable path for AI drug discovery — papers that do not share their pipelines do not produce drugs, they produce noise.

> In parallel, I developed JV Arbitre, an independent LLM audit framework with contamination detection for public benchmarks (MMLU, HumanEval, GSM8K). Current leaderboards contain up to 40% of questions seen during training by the top-ranked models — a bias that invalidates the comparisons used for purchasing decisions. JV Arbitre produces defensible scores in front of a CTO or a VC.

> Available for:

> — freelance AI drug discovery missions (2-3 days/week, daily rate €650 excluding tax)

> — independent LLM audits (3-day package, €1,200 excluding tax)

> — early-stage biotech collaborations

Sole proprietorship. Scientific direction, pipeline engineering, INPI filings, publications.

• Multi-target virtual screening pipeline: 6,500+ molecules generated, ADMET filtered (pkCSM / SwissADME), docked (AutoDock Vina) on EGFR, quinazoline scaffolds, neurological targets.

• Protein design architecture: ESM-2 / ESM-IF1 / ProteinMPNN scoring coupled with Roseta ddg_monomer and ddg_dimer for quantum biosensor design (ODMR, smURFP / miRFP670nano fluorescent proteins).

• IP portfolio: several patents filed with INPI (medical chemistry, biotech, AI).

• Zenodo Publications: 8 scientific preprints with DOI (computational chemistry, quantum biosensors, LLM benchmarking).