Darshan Deshpande
Member of Technical Staff, Research
Darshan is a research scientist at Patronus AI working on text-based world models that let agents train on simulated experience, and he has coauthored TRAIL, MEMTRACK and the reward hack detection benchmark. Before Patronus, he did research on argumentation and logical fallacy detection at USC Information Sciences Institute.
At Patronus· Newest first
Newest first
Papers
GlobeBench
A benchmark for whether text world models keep track of state, rules and side effects.
FigmaTrace
Human design workflows in Figma, captured as training data for creative design tasks.
World Modeling for Agentic RL
A language model that stands in for an environment while an agent trains.
DETOUR
A two-agent benchmark for tip-of-the-tongue search across several turns.
TRACE
54 kinds of reward exploit in code RL, and how well models catch them.
Generative Simulators
Environments that generate their own tasks, world dynamics and rewards.
MEMTRACK
Testing whether agents keep track of facts spread across workplace tools.
TRAIL
A benchmark for finding the step where an agent went wrong.
BLUR
573 tip-of-the-tongue questions that need search, reasoning and tool use.
GLIDER
A small, explainable judge model for grading LLM interactions.
Posts
Introducing FigmaTrace: A Comprehensive Training Dataset for Figma Design Workflows
Introducing Generative Simulators: Autonomously Scaling Environments for Agents
Introducing TRAIL: A Benchmark for Agentic Evaluation
No papers or posts here yet.