IEEE-TEMS logo
 

Podcast/Video Interviews by Stephen Ibaraki

A Chat with Aashna Garg: a Principal Applied Scientist at Microsoft on GitHub Copilot; AI researcher focused on multi-model orchestration, evaluation and making frontier-level AI affordable and trustworthy for every developer

This week, Stephen Ibaraki has an exclusive interview with Aashna Garg.

Aashna GargAashna Garg is a Principal Applied Scientist at Microsoft on GitHub Copilot, research lead behind Copilot Fusion (formerly Project HydraFusion), and AI researcher focused on multi-model orchestration, evaluation and making frontier-level AI affordable and trustworthy for every developer.

Aashna Garg is a Principal Applied Scientist at Microsoft, working on GitHub Copilot, the world's most widely adopted AI developer tool. She led the research behind Copilot Fusion, formerly known as Project HydraFusion, which was announced at GitHub Universe 2026 and is now generally available in Visual Studio Code, the GitHub Copilot CLI, the GitHub Copilot app and JetBrains IDEs. Her work sits at the intersection of large language models, multi-model orchestration and rigorous evaluation, with a focus on turning frontier AI research into products that millions of developers rely on every day.

Copilot Fusion represents a shift in how developers use AI: instead of choosing a single model, Copilot assembles a small team of models for each task. Developers choose an effort level, and Fusion decides which models to use, how they collaborate and how much reasoning each one applies. It uses patterns such as Cascade, where an efficient model drafts, a judge checks the result and a stronger model is called only when needed, and Critique, where models from different families review a draft in parallel before it is revised. In independent evaluation by Artificial Analysis, Copilot Fusion Max ranked #1 on Terminal-Bench 4.0 and came within two points of the top coding agent overall, at roughly half the cost, half the tokens and about 40% less time per task. Aashna introduced the approach publicly in her GitHub blog post, "Project HydraFusion: Frontier quality via multi-model orchestration."

Before Fusion, Aashna's research on HyDRA (Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools) established the routing foundation behind Auto mode in Visual Studio Code Chat, which automatically selects the right model for each request. The work was published at EMNLP 2026, one of the leading conferences in natural language processing, and has been deployed to all users of Auto mode. HyDRA was featured on the GitHub blog in "Getting more from each token: How Copilot improves context handling and model routing," which Aashna co-authored. Across these efforts, her central theme has been evaluation: building the measurement systems that show when AI is reliable, where it fails and how to make it better. She has also used Fusion to accelerate its own development, pointing the system at its own experiments to analyze results and propose improvements, an early, human-guided example of AI helping to improve AI.

Aashna grew up in Delhi, India. She earned her undergraduate degree in engineering from Delhi Technological University and her master's degree from Stanford University, where she discovered natural language processing and committed to a career in artificial intelligence.

Aashna believes that AI only changes the world when people can afford to use it and trust what it produces. She is passionate about making near-frontier capabilities accessible to students, startups and developers everywhere, including in emerging markets, and about building AI systems that are transparent, verifiable and designed with human oversight. She advocates for treating evaluation as a core engineering discipline, for combining models rather than relying on any single one, and for keeping humans responsible for goals, quality and judgment.

Aashna is particularly interested in multi-agent and multi-model systems, AI for software engineering, evaluation and benchmarking, cost-efficient AI, verification and trustworthy AI, and the future of work for developers. She encourages students to combine AI with a field they care about deeply, and professionals to make AI part of their everyday work while measuring where it helps. Her career reflects a central belief: you don't need the most resources to do frontier work. What matters is curiosity, persistence and people who believe in you.

LINKS



TO WATCH THE VIDEO INTERVIEW, CLICK ON THIS MP4 file link