Yani Guan

Incoming Member of Technical Staff at Medra, January 2027 · Ph.D., UCLA, 2026

I build symmetry-aware neural networks for chemistry — E(3)-equivariant models over atoms and trajectories, steerable encoders over molecule images, language models that read 3D directions — along with the multi-scale simulations that feed them and the GPU work that makes them trainable.

profile.jpg
Outside the lab — cameras, Shakespeare, long drives.

Research

How the work fits together Chemistry reaches a model in three forms: papers, structure drawings and 3D atoms from multi-scale simulation. Each goes through an encoder matched to its symmetry: a language model for text, SE(2)-equivariant encoders for drawings, E(3)-equivariant networks for atoms. Irreps-to-text attention and contrastive alignment join them in one latent space with retrieval over all three; post-training turns that into a co-scientist that proposes the next simulation. GPU kernel and precision work sits under every stage. DATA Papers methods, results, claims Structure drawings figures in the literature Multi-scale simulation DFT · MD · kMC · P2D ENCODER · MATCHED SYMMETRY Language model tokens · no geometric group SE(2)-equivariant encoders CN steerable CNN · Ouroboros · VERDICT E(3)-equivariant networks polarizable FF · WignerFlow · world model SMILES → ETKDG → MACE-OFF ALIGNMENT Irreps-to-text YLM · text queries l ≥ 1 Contrastive alignment one latent space Multimodal RAG retrieval over all three POST-TRAINING & AGENT TASK + TOOL DATA 254,527 rows · leak-safe splits LoRA SFT DPO tested · SFT kept CO-SCIENTIST DeepSeek Harness · default-deny jobs proposes the next simulation — and submits the jobs that check it COMPUTE GPU Clebsch–Gordan tensor products · channel-wise (uvu) kernels · cuEquivariance · torch.compile · bf16 / tf32 · A100

Scroll the diagram sideways →

Chemistry reaches a model as papers, drawings and simulated atoms. Each gets an encoder that respects its symmetry: none for text, planar rotations for drawings, E(3) for atoms. Alignment puts them in one space. Post-training and a harness turn that space into a co-scientist whose proposals go back to simulation as jobs. GPU work sits under every stage.

Vision: a lab agent that understands chemistry

A lab agent that understands chemistry and biology A line drawing of a two-armed robot working at a lab bench, linked to a GPU, with a pre-experiment loop below. Filled marks are built; hollow marks and dashed outlines are planned. Brain: YLM reads text and 3D geometry together and recalls papers, structures and simulations from a molecular memory; strict checks that turn text into Lean proofs, SQL queries and protocol scripts are planned. Eyes: Ouroboros OCSR turns paper figures into structures, and four models vote to flag errors; spectra, gels and a lab camera are planned. Ears, planned: chemistry and biology protocols, scientist goals by voice or chat. Touch, planned: tactile pads on a gripper feel the last centimeter where glass and liquids fool cameras, for tip seating and cap grip. Hands, planned: one arm holds a pipette over a well plate, the other a sample tube above a tube rack; pipetting, tip-to-well alignment and liquid-handler scripts. First principles: quantum chemistry as ground truth and multiscale models from electrons to cells; distilling them into small, fast students is planned. Physics intuition: WignerFlow learns MD frame by frame (shown so far on argon), and force fields are gated by past failures; simulating each step before running is planned. Body: a bounded harness diagnoses failures, catches repeats and puts a default-deny gate on every job. Action planning, planned: protocol to steps to motion instead of hard-coded moves. Compute, planned: cloud GPUs only in short bursts and small quantized models on the edge. Before every experiment: recall molecular memory and past failures (built); rehearse in a GPU digital twin of the bench, an Isaac-style physics sim (planned) and with learned MD (built, argon so far); verify with Lean, SQL and protocol checks (planned) behind the default-deny job gate (built); run the real experiment, guided by camera and touch (planned); write results and failures back to memory, which trains the Decision API (planned). A planned Decision API closes the loop: trained on our own lab data, it grows out from a trusted zone and asks why, not who is right. GPU BRAIN YLM Reads text and 3D geometry together Recalls papers, structures, simulations Strict checks: Lean · SQL · protocol EYES Ouroboros OCSR Turns paper figures into structures Four models vote to flag errors Spectra, gels and a lab camera EARS Lab language Chemistry and biology protocols Scientist goals by voice or chat HANDS Instrument control Pipetting, tip-to-well alignment Writes liquid-handler scripts TOUCH Tactile sensing Feels where glass fools cameras Force-guided tip seating, cap grip FIRST PRINCIPLES DFT teacher Quantum chemistry as ground truth Multiscale: electrons to cells Distilled into small, fast students PHYSICS INTUITION WignerFlow Learned MD, frame by frame Force fields gated by past failures Simulates each step before running BODY Bounded harness Diagnoses failures, catches repeats Default-deny gate on every job ACTION PLANNING Step planner Protocol to steps to motion Replaces hard-coded moves COMPUTE Sim-to-real compute Cloud GPUs only in short bursts Small quantized models on the edge BEFORE EVERY EXPERIMENT Recall Molecular memory Past failures as gates Rehearse GPU digital twin (Isaac-style) Learned MD (argon so far) Verify Lean · SQL · protocol checks Default-deny job gate Run Real experiment at the bench Guided by camera and touch Write back Results, failures to memory Trains the Decision API Decision API trained on our own lab data · grows out from a trusted zone · asks why, not who is right built planned future part

Scroll the diagram sideways →

Today's self-driving labs mostly train an athlete to drill the same moves. A lab agent that understands the chemistry behind each protocol step could plan better and find strategies people don't expect.

Planned: a recursive self-improvement loop

Recursive self-improvement loop (planned) A quiet field of faint gray nodes and edges, with one highlighted path in the accent color tracing a cycle: generate new data, verify it with the Lean checker, SQL checks, protocol checks and simulation, keep only what passes, train on it, expand the trusted regime where bench results confirm it, then generate again. Planned. RECURSIVE SELF-IMPROVEMENT — PLANNED Generate new data Verify Lean · SQL · protocol · sim Keep verified only Train SFT · GRPO on what passed Expand truth regime where bench confirms

Scroll the diagram sideways →

Planned. The agent writes new data, keeps only steps that pass the Lean checker, SQL and protocol checks and simulation, trains on what it kept, and widens the trusted regime only where bench results confirm it — then repeats. A human signs off each round.

I work on machine learning for chemistry: equivariant models, chemistry vision and language models, and agents that run simulations.

VERDICT
Four different structure recognizers vote. When they disagree, the structure is flagged as wrong, with no labels needed. AUROC 0.916. Chemists rated 400 released labels at 0.995 precision. arXiv:2608.22183
Image-to-SMILES post-training
LoRA SFT on three open VLMs reached 59.8% exact match on 331 journal figures, against 58.3% for MolScribe. DPO pushed preference accuracy to 90–97% while exact match fell to 0–2%, so I kept SFT.
Electrolyte co-scientist
Qwen3.6-27B with LoRA on 254,527 task, evidence and tool-use rows. Each MD job is checked before it is submitted, and repeat jobs are caught.
An autoregressive E(3)-equivariant transformer that generates MD trajectories. Public results so far are on argon.
A CN steerable CNN as the image encoder for structure recognition. It handles rotations but not reflections, because a mirror image flips stereocenters.
Text tokens attend to l ≥ 1 irreps through Clebsch–Gordan products. Equivariance error is below 10−9 in fp64.

Agents for science

Failure-gated force-field harness
Recorded failure modes of past polarizable force-field fits become automated gates; a candidate field that repeats a known short-range collapse or typing break is rejected before use.
Consensus labels for self-training
Independent structure readers vote on literature figures; only agreed labels enter training, and real crops, not more synthetic drawings, moved accuracy.
Multimodal molecular retrieval
A molecular knowledge graph keyed on InChIKey, with identity, structure-similarity and text search linking literature, structures and simulations.
Bounded scientist agents
Tool-using electrolyte and cell scientist agents that inspect trajectories, diagnose failures and prepare jobs; submission is default-deny and model text never runs as a shell command.
Cell-scale simulation
Large P2D and 3D electrothermal cell simulation campaigns for design studies.

News

Nov 19, 2026 Excited to share that I will join Medra as a Member of Technical Staff in January 2027.
Nov 16, 2026 Defended my Ph.D. at UCLA. Download the slides (PDF)
Nov 03, 2026 Google Antigravity Selected for six months of upgraded Google Antigravity access for scientific research.
Oct 01, 2026 Started three equivariant-network projects for chemistry: WignerFlow iconWignerFlow, YLM iconYLM, and Ouroboros iconOuroboros.
Jun 08, 2026 Excited to join SES AI Corp as an Applied Scientist Intern for the summer!

All news

Experience

2026
SES AI Applied Scientist Intern
Post-training and evals for chemistry vision and language models, multi-scale electrolyte and cell simulation, and co-scientist agents that plan and run jobs. Notes: Overview · Molecules · Electrolytes · Cells.
2022 — present
UCLA Ph.D. Research
DFT and multi-scale modeling of electrochemical interfaces, electrode degradation, and reaction selectivity; LLM- and vision-based tooling for computational chemistry.
2022
DP Technology Algorithm Researcher Intern
Production ML pipelines for neural network potentials and scalable scientific computing workflows; developer-community and documentation work in the DeePMD ecosystem.
2018 — 2022
Hebei University of Technology Undergraduate Research
Multi-scale simulation (DFT–kMC), machine learning for catalysis, and electrochemical experiments on bifunctional catalysts for Li–S and Zn–air batteries; three first-author publications.

Selected publications

  1. OCSR VLM
    ocsr_real_data.png
    Real Data Closes Synthetic-to-Real Gap in Optical Chemical Structure Recognition
    Yani Guan, Dengpan Dong, Zi Wei, and 4 more authors
    arXiv preprint arXiv:2608.09100, 2026
    * Equal contribution. Vision-language models for image-to-SMILES: real labeled depictions, not more synthetic data, are what close the synthetic-to-real gap.
  1. VERDICT
    VERDICT: Agreement Beats Pixel-Space Verification in Real-Document OCSR
    Yani Guan, Dengpan Dong, Shuang Luo, and 6 more authors
    arXiv preprint arXiv:2608.22183, 2026
    Label-free verification for image-to-SMILES at scale: agreement among four architecturally distinct recognizers reaches AUROC 0.916, while re-rendering a prediction and comparing pixels barely beats chance at 0.547.
  1. MU-PFF MD
    mu_md_platform.png
    High-throughput Molecular Dynamics Simulation on an AI-powered Platform
    Dengpan Dong, Yani Guan, Shuang Luo, and 5 more authors
    ChemRxiv preprint, 2026
    MU-PFF: a machine-learned many-body polarizable force field plus automated system building and trajectory analysis, taking an electrolyte formulation from SMILES to predicted properties without manual parameterization.
  1. Nature Catal.
    Desolvated cations promote CO2 electroreduction through partial covalent interactions
    Yu Shan, Maximilian Jaugstetter, Julian Feijóo, and 14 more authors
    Nature Catalysis, Sep 2026
    Yu Shan, Maximilian Jaugstetter, Julian Feijóo and Yani Guan contributed equally.
  1. Ag Amination
    ag_amination.jpeg
    Electrochemical Amination of Acetone using Ag as Cathode: Mechanism and Role of Pb Impurities for Hydrogen Transfer
    Yani Guan, Justus Kümper, Angelina Cuomo, and 6 more authors
    Catalysis Science & Technology, 2026

All publications