Research · products · teaching
I am Boxi Yu, a software engineering researcher at Lero. I study how to test and evaluate coding agents and AI systems so that people can decide when to trust them, then turn those lessons into tools people can use and courses people can learn from.
Strengthening tests and examining the reliability of benchmark results.
Using relations between programs and executions to detect inconsistencies.
Controlled transformations reveal errors in vision and language systems.
LightAD, OpenRCA 2.0, and CLEANet examine model efficiency, causal diagnosis, and robustness to contaminated data.
Co-founder
Alongside my research I co-founded OneAI and Pacific Studio, where we build AI products that keep people in the decision seat.
ONEAI
Turn a learning objective into a question set, rubric, and student-level feedback loop. Teachers stay in the decision seat; AI handles the first pass.
Co-founder · build, prove, understand
Three products from Pacific Studio, which I co-founded, show how AI can support the software lifecycle: ship the work, check the proof, and read the ideas that shape it.
Build
Issue-driven orchestration for the full software lifecycle, with Claude Code, Codex, and Gemini CLI working across design, development, testing, and deployment.
Prove
AI-assisted formal verification that turns plain code into machine-checked proof goals, with an offline workflow for C, Java, Dafny, and Rust.
Understand
An interactive paper reader that combines layout-aware parsing, sentence-level highlights, bilingual reading, and answers that point back to the source.
Agents generate. You verify, decide and explain.
A practical twelve-week course for building with coding agents while keeping specifications, tests, review, security, and judgment in the loop.
I have also mentored eleven students, nine of whom are now Ph.D. students. See where they went.
BlogWhy evidence belongs in the workflow
A product should make the important evidence easy to inspect, from a student hint to a coding-agent patch.
Four new preprints: on RAG hallucination detection, conversational agent testing, code specification alignment, and causal root-cause analysis. All are on the publication list.
Our ICML and ICML Position papers were accepted: "SWE-ABS: Adversarial Benchmark Strengthening Exposes Inflated Success Rates on Test-based Benchmark" and "How Should We Build A Benchmark? Revisiting 274 Code-Related Benchmarks For LLMs".
Our paper "UTBoost: Rigorous Evaluation of Coding Agents on SWE-Bench" was accepted by ACL'2025.