Chen Qian

I am an AI researcher and engineer focused on building robust AI systems that improve themselves from sparse human feedback. My research spans post-training, AI system evaluation, and in-context learning. My goal is to enable deploying high-quality AI systems across all industries, not just tech. Most of my work is open source and available on my GitHub.

I am a PhD student at CMU SCS (since August 2026), working on self-improving AI systems and AI for software engineering (AI4SE).

I am also a Member of Technical Staff (Research Engineer) at Unconventional AI. Previously, I spent 7+ years at Google and Databricks, working mainly on open source projects, including Keras, DSPy, and MLflow.

Research Interests

• Self-improving AI systems with sparse human feedback: Systems that keep getting better from their own traces and from scarce, noisy, or delayed human signals, including automatic prompt and weight optimization, self-refinement loops, and continual improvement of compound AI pipelines with minimal human supervision.
• AI for software engineering (AI4SE): Reliable AI systems for real-world software engineering tasks, including code generation, debugging, refactoring, and end-to-end agentic development.
• AI system evaluation: Benchmarks and evaluation frameworks for domain-specific applications (hardware design, scientific reasoning, finance), with a goal of minimizing human effort during the evaluation process.
• Compound AI systems: Architecture for AI agents and enhancing retrieval quality in AI systems.
• Efficient training and inference: Reducing communication overhead in distributed systems and new model/layer architectures for computational efficiency.