Hi, I'm Weiran Yao

I'm a Research Scientist & Manager
at Salesforce AI Research
I develop multi-agent AI systems and train their language models.
At Salesforce, I work on AI agent systems including SlackAgents, AgentLite, CodeGenie, SWE Agents, and WebAgent.
For models, I've led post-training of Retroformer, a genertive critic model for self-reflection. I've worked on xLAM, a family of state-of-the-art language models for function calling. I conducted research on synthetic data generation pipeline for APIGen and AgentOhana.
For products, I've led delevopment of AIOps for cloud incident causation analysis, enhancing operational efficiency on Salesforce Availability Cloud. I worked on function call and structured output API endpoints of Salesforce xLAM services.
I did my Ph.D. and M.S. in Machine Learning at CMU, focusing on model interpretability, where I was advised by Dr. Kun Zhang.

Featured Research Publications

Latest research for fans of AI Agent, RL and Interpretability.

Generative Critic Model Optimized for Self-Reflection and Reasoning Capabilities of AI Agents
ICLR 2024
World's Best Open-Source Model for Function-Calling
arXiv 2024
Salesforce's In-House Library for Multi-Agent Orchestration and Reasoning
arXiv 2024
AI Software Engineer with 55% resolve rate on SWE-Bench Lite
ICLR 2025
Optimizing Principled Reasoning and Acting of LLM Agent
CoNLL 2024
Salesforce Library for for Fast and Scalable Causal Reasoning
arXiv 2023


Featured Engineering Projects

Innovative AI systems I've developed and deployed at scale.

Scalable Collaboration for Multiple AI Agents in Workspaces
Jan 2024 – Present
Enhancing IDE Productivity through AI Code Planning, Editing, and Execution
Aug 2024 – Present
AIOps Augments SREs' Capabilities for Automating Operations
Jan 2023 – Dec 2023


Featured Talks & Presentations

My favorite talks and presentations given at industry AI conferences and workshops.

Witness How xLAM and Multi-Agent Framework Revolutionize Automation with A Live Demo for Marketing Sales
Dreamforce 2024
Explore Power of Multi-Agent Reasoning with Software Engineering Agents of Diverse Strengths
CAMEL-AI Workshop 2024