Bird's-eye view

Research map

Select a topic to jump directly to the project.

Research programReliable collaborative AI

Research portfolio

Current research

01Research program

Measuring Multi-Agent Scientific Reasoning

AgentsSci

Open, schema-stable resources for studying how AI agents collaborate on scientific problems—from protocol choice and cost to critique uptake, trajectory value, and agent attribution.

  • 7,467 problems
  • 82,224 trajectories
  • 5 protocols
02Decision

When does collaboration pay?

Cost-Aware Protocol Routing

Studies whether a system can decide when collaboration is worth the added inference cost and which protocol should be used for a given problem.

  • Routing
  • Cost
  • Failure risk
03Mechanism

Precise but uncoupled

Critique Uptake

Separates reviewer precision from critique uptake to explain why technically correct feedback may still fail to improve a solver’s reasoning.

  • Reviewers
  • Solvers
  • Communication
04Mechanism

Wrong but useful

Trajectory Value

Measures whether an intermediate message improves downstream team reasoning, extending value beyond the correctness of the message itself.

  • Messages
  • DHD
  • Learning labels
05Learning

Who gets the reward—and the blame?

Evaluation-Aligned Credit Assignment

Transforms system-level evaluation into signed, credit-conserving signals for individual agents and messages.

  • Attribution
  • Process rewards
  • Training signals

Research portfolio

Earlier work

06Ph.D. research

Generative and multi-fidelity scientific AI

AI for Materials & Molecular Discovery

My Ph.D. work connected generative modeling, molecular language models, reinforcement learning, and multi-fidelity learning to materials and molecular design—from photovoltaic microstructures to organic chemical space.

  • Molecular LMs
  • Multi-fidelity
  • Inverse design
  • RL
07Multimodal AI

AI-ready biodiversity at scale

BioTrove

A NeurIPS 2024 Spotlight project presenting 161.9 million research-grade images across approximately 366,600 species, along with 40 million captioned training examples, multimodal models, and biodiversity benchmarks.

  • 161.9M images
  • 366.6K species
  • CLIP
  • Spotlight
08Industry research

Visiting Scientist at Corteva Agriscience

Preference-Based Post-Training

Applied RLHF with preference-based insights for domain-specific model customization. Additional technical details will be added after public wording is confirmed.

  • RLHF
  • Preference insights
  • Domain adaptation
  • Scientific AI
09Industry research

AI Resident at X, Google LLC

Multi-Agent Systems at Google X

Developed multi-agent collaboration frameworks using a Mixture-of-Experts architecture with multimodal data streams. More project detail will be added after public wording is confirmed.

  • Multi-agent systems
  • Mixture of Experts
  • Multimodal
  • Collaboration