Portrait of Jianfeng Gao

Jianfeng Gao

Technical Fellow & Corporate Vice President

News & features

Benchmark overview showing ten spatial reasoning and planning tasks grouped into two rows. The top row, labeled “Reasoning,” includes Maze, Assembly, Bead, Sheep, and Knot. The bottom row, labeled “Planning,” includes Pipe, One Stroke, Swap, Chat Noir, and Untangle. The MindTopo logo and title are centered between the two categories.
Microsoft Research Blog

| Yunfei Ge, Anbang Liu, Qineng Wang, Johnalbert Garnica, Zihan Wang, Reuben Tan, Jianfeng Gao, Ruohan Zhang, Yining Hong, Jiajun Wu, and Manling Li

A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning and planning.

Three Orchard framework components with benchmark results
Microsoft Research Blog

| Baolin Peng, Wenlin Yao, Qianhui Wu, Hao Cheng, and Jianfeng Gao

Orchard is an open-source framework for the research community to train and evaluate AI agents across task types. It reduces complexity while supporting strong performance from smaller models by enabling researchers to reuse the same infrastructure.

Figure 1. EvoLib transforms raw experiences into reusable skills and insights, then continually evolves them through consolidation and dynamic weighting.
Microsoft Research Blog

| Weijia Xu, Alessandro Sordoni, Zelalem Gero, Michel Galley, Eric Yuan, and Jianfeng Gao

LLMs do not get smarter just by remembering more. EvoLib turns experience into evolving knowledge, taking reusable skills and insights that help models learn and adapt across tasks long after deployment.

Flint blog | three white line icons on an abstract green background; bar chart icon, connected nodes icon, flowchart icon
Microsoft Research Blog

| Chenglong Wang, Alper Sarikaya, Scott Tsukamaki, Michel Galley, and Jianfeng Gao

Short chart specifications are easy to write, but often produce uninspiring results. Flint is an open-source visualization language that offers a middle path, letting AI agents create expressive charts from compact, human-editable specifications.

Understanding the brain | four white line icons on an abstract purple background: brain icon, chat bubble icon, circle with a checkmark icon, search icon
Microsoft Research Blog

| Chandan Singh and Jianfeng Gao

Researchers introduce generative causal testing, which translates black box models into clear hypotheses and verifies them in the scanner, revealing what specific brain regions respond to in language.

Three minimalist white line icons on a textured blue‑green gradient background: a rising bar chart on the left, a central hub‑and‑spoke network diagram in the middle, and a checkmark inside a circle on the right.
Microsoft Research Blog

| Chenglong Wang, Scott Tsukamaki, Michel Galley, and Jianfeng Gao

Data Formulator introduces AI-powered analytics for enterprise data workflows. Data teams can easily bring enterprise data into an AI-ready workspace where users can explore, analyze, and visualize data with AI agents to turn raw data into actionable insights.

AsgardBench | three whit icons on a blue to purple gradient background | first icon shows a laptop screen with a eye in the upper right corner, second icon shows relational nodes | third icon is a security shield with a checkmark
Microsoft Research Blog

| Andrea Tupini, Lars Liden, Reuben Tan, Yu Wang, and Jianfeng Gao

Imagine a robot tasked with cleaning a kitchen. It needs to observe its environment, decide what to do, and adjust when things don't go as expected, for example, when the mug it was tasked to wash is already clean, or…

V2GP framework | Three white line icons, showing a target within a rounded square, a checklist, and a robotic arm, on a blue‑to‑green gradient background.
Microsoft Research Blog

| Sehun Jung, HyunJee Song, Dong-Hee Kim, Reuben Tan, Jianfeng Gao, Yong Jae Lee, and Donghyun Kim

Vision-language models (VLMs) use images and text to plan robot actions, but they still struggle to decide what actions to take and where to take them. Most systems split these decisions into two steps: a VLM generates a plan in…

blue and purple gradient background with decorative white icons
Microsoft Research Blog

| Ke Yang, Michel Galley, Chenglong Wang, Jianfeng Gao, Jiawei Han, and ChengXiang Zhai

It seems counterintuitive: giving AI agents more memory can make them less effective. As interaction logs accumulate, they grow large, fill with irrelevant content, and become increasingly difficult to use. More memory means that agents must search through larger volumes of…