Rahul Pandita

Rahul Pandita is a staff researcher at GitHub Next whose work spans Copilot Voice, Copilot Workspace, and experimental coding agents. He designs AI systems that help developers express intent, understand unfamiliar software, and verify proposed changes while retaining control over their work.

Pandita earned a doctorate in computer science from North Carolina State University in 2015, researching ways to extract semantic information from the natural language surrounding software. After postdoctoral work at NC State, he joined Phase Change Software in 2016 and became a senior research scientist focused on program comprehension and human-computer interaction. There, he led development of MIA, a conversational assistant for understanding legacy software, and studied how developers manage context and respond to cognitive bias. His academic profile documents additional research in API specifications, mobile security, and automated testing.

He joined GitHub Next in December 2021 and led the team behind Copilot Voice, which enabled developers to navigate editors and manipulate code through speech. The project informed capabilities associated with Copilot Chat and VS Code Speech, making programming more accessible to people for whom sustained keyboard and mouse use is difficult. Pandita has credited the original project team, while Microsoft’s account of voice-driven coding describes its accessibility implications.

As a founding member of the Copilot Workspace team, Pandita worked on the runtime infrastructure connecting natural-language tasks to editable specifications, implementation plans, code changes, and executable previews. His AI Engineer demonstration shows this workflow adding Rust syntax highlighting to an existing website: the developer can revise the specification, correct the plan, inspect changes, run the application, and review the result.

• Developer agency and verification. Pandita treats editable plans, executable checks, and human review as product requirements. His coauthored research on developer trust examines how output quality, practical usefulness, preserved intent, and cognitive style shape adoption of generative coding tools.

• Copilot Next Edit Suggestions. Instead of predicting only what belongs at the cursor, this approach anticipates related changes elsewhere: adding a parameter can require updating function definitions, arguments, and documentation. Relevant edits may also sit outside the visible editor or in unopened files.

• Extract, Edit, Apply. This GitHub Next research project treats code as the durable artifact and natural-language specifications as temporary, editable representations that developers can use to generate and inspect proposed changes.

• Discovery Agent. Pandita led development of a repository setup and testing prototype that resolves dependencies, builds projects, and runs tests in containerized environments, giving coding agents a practical way to validate their work.

His independent projects include react-term, a React and React Native terminal emulator using shared memory, Web Workers, and GPU rendering, alongside research into agent governance and security covering agent configuration, skills, plugins, and Model Context Protocol integrations.