As a Research Scientist with a deep expertise in chemoinformatics and intelligent chemical design, I focus on advancing the field through innovative research in small molecule reactivity and cheminformatics methods for AI models. With a solid foundation in medical and medicinal chemistry, I've contributed to diverse projects ranging from molecular co-crystals research to receptor/ligand interactions analysis, utilizing advanced computational tools and algorithms.
For a detailed overview of my education, work experience, research activities, personal qualities, and skills, please check my resume.
Projects
Matcha is a molecular docking pipeline based on multi-stage Riemannian flow matching. It connects learned pose generation with physical-validity filtering, GNINA minimization/scoring, CLI inference, batching, and multi-GPU workflows.
Stack: Python, PyTorch, RDKit, GNINA, PoseBusters, ESM, Typer, Multi-GPU
Bento is a reproducible benchmark for protein-ligand docking methods. It supports curated datasets, ligand annotations, raw prediction artifacts, pocket-similarity workflows, tests, documentation, and HPC/SLURM execution.
Stack: Python, RDKit, uv, Pytest, SLURM, Docking Benchmarks, Pocket Similarity
HEDGEHOG is a staged evaluation pipeline for generative molecular design. It runs molecule preparation, descriptor calculation, structural and medchem filters, retrosynthesis checks, docking, pose-quality filtering, and final reports.
Stack: Python, RDKit, AiZynthFinder, GNINA, SMINA, Matcha, CLI/TUI, HTML Reports
SpectralixBenchmark is a chemistry benchmark and evaluation package for agentic language models. It evaluates reaction understanding, single-step retrosynthesis, route-level synthesis planning, MS/MS-related tasks, deterministic scoring, and rubric-based LLM judging.
Stack: Python, OpenAI-compatible APIs, PydanticAI, OpenShell, RDKit, LLM-as-Judge
posecheck-fast is a high-throughput docking pose evaluation package with symmetry-corrected RMSD and lightweight PoseBusters-style distance and clash filters, designed for fast validation inside larger docking and model-evaluation loops.
Stack: Python, RDKit, spyrmsd, PyTorch, PoseBusters, Pytest, Ruff
Burette is a macOS molecular preview tool with Finder and Quick Look integration. It supports Mol* 3D visualization, fast XYZ/CUBE rendering, RDKit-powered compound grids, SMARTS filtering, search, sorting, and export workflows.
Stack: Swift, macOS, Quick Look, WebKit, Mol*, RDKit.js, TypeScript