Kemal Özkırşehirli

Kemal Özkırşehirli

- International student from İstanbul, Türkiye · B.Sc. candidate at MIT · GPA 5.0/5.0 · rare triple major in Computer Science and Engineering: AI/ML, Physics: Chemical Physics, and Mathematics: Mathematical Biology · minors in Philosophy and Writing · expected May 2027 · graduating one year early

- Principal Investigator, Özkırşehirli Group

- Research across MIT Chemistry, MIT CSAIL, MIT Physics, A*STAR EDDC, PRISM/W2D2, and Harvard Medical School

- Teaching Fellow, MIT EECS

- Cambridge, MA, USA - kemalozk@mit.edu

Awards:

Medalist, 56th International Chemistry Olympiad
Medalist, 58th International Mendeleev Chemistry Olympiad
Silver Medal, TÜBİTAK National Chemistry Olympiad
Columbia Science Research Fellow Named Scholar
World Science Scholar
AIME qualification and AMC Hall of Fame recognition
AlgoTrade Hackathon 2026 selectee

My work and the Özkırşehirli Group aim to make difficult scientific systems parsable, computable, verifiable, and decision-ready. (1) We develop MeshAnyOrder, an order-agnostic autoregressive transformer for topology-aware 3D mesh generation from point clouds. (2) We build multi-stage CADD pipelines for TBXT/brachyury, integrating docking, QSAR, CNN rescoring, co-folding, free-energy scaffolding, selectivity, and sourceability. (3) Scientific-software directions include VeriQSM + QSMBench for artifact-grounded verification of scientific false success, Kadanoff-GNN-RG for learned coarse-graining and phase analysis, Kupcinet–Getz Reaction–Diffusion AI for solver-faithful benchmarking, DFT → kMC for auditable multiscale kinetics, and a structure-informed antibody-RL scaffold. (4) I develop and evaluate generative, foundation, and agentic systems for protein design, chromatin organization, autoimmune target discovery, molecular simulation, and robust scientific decision-making.

Professional

About me - Professional - Education - Recognition - Activities - Contact - Resume

Research - MeshAnyOrder - TBXT / CADD - Protein Safety - Chromatin - Target Discovery - Protein Design - VeriQSM - Antibody RL - Kadanoff-GNN-RG - Reaction–Diffusion AI - DFT → kMC

Teaching - 6.1200[J] - Renormalization Group

Writing - About - Collection of Prose and Poems

Skills - Computational Stack - CADD - Languages - Interests

Profiles - Google Scholar - GitHub - LinkedIn

Community - AI@MIT - MIT BioMakers - Schwarzman Advisor - MIT Student Council - ClubChem - MIT Gala

AI/ML for Science - Selected Work

Molecular Simulation, CADD, and Chemical Physics

TBXT - 11-person chordoma/TBXT hit-identification project targeting PDB 6F59 site F; 2,274 prior-art compounds, 737 analogs, 30,000 BRICS recombinations, a 570-compound novelty-filtered pool, and a docking/QSAR/GNINA/Boltz-2/MMGBSA-FEP/selectivity funnel

VeriQSM + QSMBench - verification-first scientific-agent framework for quantum chemistry and statistical mechanics: typed workflow IR, static and physics verifiers, bounded repair/refusal policies, provenance ledgers, PySCF-compatible calculations, trajectory/stat-mech analyses, and explicit separation of nominal completion from scientifically verified success

Kadanoff-GNN-RG - hierarchical-lattice renormalization-group framework for random-bond spin-s Ising models; direct continuum RG operators, finite-spin distribution comparison, symmetry-aware features, Wasserstein/MMD-style distances, and uncertainty-aware phase-topology analysis

Buchwald Lab - CuH-catalyst development for asymmetric methylation and alkylation of vinyl boronate esters, with 1H-NMR, chiral-HPLC, substrate-scope expansion, and multi-gram scale-up

Pedal AI - Python/LangGraph and FastAPI infrastructure for chemistry agents, molecular-prediction engines, inspectable retrosynthesis trees, provenance, and decision-ready synthesis workflows

Kupcinet–Getz Reaction–Diffusion AI - reaction–diffusion scientific-computing framework for nonlinear oscillators, traveling waves, and synthetic morphogenesis; deterministic ODE solvers, chemical-Langevin dynamics, Gillespie SSA, spatial RDME, convergence audits, calibration, and abstention

Sepal AI / Mercor - Graduate-level chemistry LLM data, retrosynthesis and reaction-mechanism tasks, chemical-validity rubrics, prompt analysis, and failure-mode annotation

DFT → kMC - mechanism-aware decision-stability pipeline linking Gaussian/ORCA thermochemistry to Eyring rates, ODE/CME/Gillespie kinetics, solvent/pathway/product ranking, uncertainty checks, and provenance-rich scientific reports

Coursework and Toolchain

Stack - Python, C/C++, CUDA, PyTorch, Transformers, GNNs, HPC/GPU programming, Linux, Bash, Git, SLURM/PBS/qsub, RDKit, OpenMM/OpenFE, PySCF, LangGraph, FastAPI, and LaTeX

Chem/Physics/Simulation - 8.592[J] Statistical Physics in Biology; 2.094[J] Atomistic Modeling and Simulation of Materials and Structures; 8.05 Quantum Physics II; 5.601 Thermodynamics I; 5.351 Fundamentals of Spectroscopy; 5.43 Advanced Organic Chemistry

CS/AI/Systems - 6.7960 Deep Learning; 6.5060 Algorithm Engineering; 6.4610 Natural Language Processing; 6.3930 AI and Decision Making in Medicine; 6.1220[J] Advanced Algorithm Design-Analysis; 18.404[J] Theory of Computation; 18.615 Stochastic Processes; 6.5240 Sublinear Time Algorithms; 6.1910 Computer Structures; 6.1903 C and Assembly; 6.100B Computational Thinking; 6.1200[J] Discrete Mathematics; 18.600 Probability; 18.06 Linear Algebra

Humanitas Core - 24.245 Theory of Models; 21W.747 Rhetoric; 21W.745 Advanced Essay Workshop; 21W.762 Poetry Workshop; 24.251 Philosophy of Language; 24.05 Philosophy of Religion; 17.01[J]/24.04[J] Justice; 24.121 Metaphysics; 21W.738[J] Memoir

Biomolecular AI, Protein, and Genome Modeling

MeshAnyOrder - order-agnostic autoregressive transformer for point-cloud-conditioned 3D mesh generation with quantized face tokens, arbitrary traversal seeds, 3D RoPE, triangle/quad tokenization, topology-aware constraints, frontier-parallel decoding, and publication-grade ablations

PRISM - Adversarial evaluation of protein foundation models across biological plausibility, sequence-structure-function consistency, developability, uncertainty calibration, claimed mechanisms, and failure taxonomies

ChromoGen V2 - conditioned 3D genome generation and evaluation pipeline for single cells; sequence windows, perturbation labels, regulatory readouts, coordinate ensembles, contact/P(s)/polymer realism, condition separation, calibration, leakage checks, and attribution diagnostics

V2M - provenance-first variant-to-mechanism evidence graph for IBD/autoimmune target discovery, connecting disease, locus, QTL/colocalization, effector gene, cell context, mechanism, and target-control tiers before foundation-model extensions

EVEdesign - Open-source, method-independent protein design across a 21-person, 18-institution, 8-country collaboration, with uncertainty-aware multi-objective candidate selection and lab-in-the-loop workflows

Antibody RL - risk-constrained, structure-guided policy-optimization scaffold for antibody sequence research; ESM-2-compatible antigen embeddings, structure-informed cross-attention, supervised warm starts, selective PPO, uncertainty penalties, and evidence-lineage controls

Education and Recognition

MIT - GPA 5.0/5.0; rare triple major in Computer Science and Engineering: AI/ML, Physics: Chemical Physics, and Mathematics: Mathematical Biology; minors in Philosophy and Writing; expected May 2027; graduating one year early

Columbia / MIT Transfer - Columbia Science Research Fellow Named Scholar with Likely Letter and full-ride early award; transferred to MIT through a 22-of-764 selection process

Olympiads - IChO medalist, Mendeleev medalist, TÜBİTAK national silver medalist, and Istanbul Science Olympiad chemistry gold medalist

WSS / AlgoTrade - World Science Scholar 2023 cohort and AlgoTrade Hackathon 2026 selectee from 1,000+ applicants across 51 countries and 148 universities

AMC / Waterloo - AIME qualifier and AMC Hall of Fame recognition; CHEM13 global 2nd, Galois international distinction, and Sir Isaac Newton Book Prize