Research
I work on systems that turn scientific judgment under uncertainty into replayable environments, verifiers, rewards, and model-improvement loops.
Current threads
Scientific autoresearch
Dynamical CLI gives scientific agents one interface for virtual campaigns and independently authorized physical experiments.
Scientific post-training
Replay environments compiled from physical experiments train a 35B open model for evidence-conditioned campaign control.
Scientific evaluation
An environment compiler and closed-loop evaluation of how scientific agents convert physical evidence into better decisions.
Test-time scaling
A persistent agent learns to operate an unfamiliar scientific instrument by drafting a procedure, running it against a simulator, and repairing it from what fails. It improves from every attempt but never grades its own work; a separate verifier the model cannot change controls what enters the skill library, and the whole loop self-verifies in simulation before any hardware is touched. Across three instrument families it acquired, repaired, and evolved skills without ever promoting an invalid one.
Verification
Probe-gradient guidance steers an unconditional crystal diffusion model toward a target band-gap window without retraining the generator. Reported gains are computational; physical qualification remains untested.
Post-training
Ordinary pretraining chunks become continuation-choice data, interleaved-thought SFT examples, and RL mid-training environments.
Interpretability
Signal detection measures whether models revise for valid critique while resisting invalid critique across architecture, domain, and source pressure.
Papers
Scientific agents / Benchmark construction
A benchmark-construction method that turns disagreement between public scientific models into verifier-backed agent tasks. The first release is 165 regulatory-genomics triage tasks where 6 frontier agents reach an 11.3% aggregate strict solve rate and none exceed 15%.
Post-training / Trajectory data
An incentive-aligned agent arena becomes a post-training trajectory substrate. A structural-quality filter turns raw ShoppingBench subnet traces into a trainable corpus that lifts Qwen3-4B from 18.0% to 42.7% strict solve rate, training on a fraction of a single day of arena output.
RLVR / Code
RL-trained 120B KernelBench policy plus verifier-filtered selection under fixed inference budgets.
Security agents / Calibration
Dual-control incident-response benchmark measuring whether agents verify evidence before containment under adversarial context, with deterministic oracle scoring and published traces.
Agents / Continual learning
ATLAS modeled the advisor/executor pattern as a continual-learning system. A stronger teacher guided a lower-cost production agent, while reward-scored trajectories were stored as persistent learning memory and reused for inference-time adaptation, RL, and on-policy distillation.