Associate LLM Engineer @ Huawei Canada Research Centre
Radin Shayanfar
I build, evaluate, and write about LLM-driven systems — with a focus on coding agents and agent evaluation.
I work on coding agents at Huawei Canada’s Research Centre. Before that, I got my master’s from Queen’s University (2025). My research bridges NLP with practical engineering to make LLM-driven systems more capable, controllable, and reliable.
at a glance
Huawei Canada · since Sep 2025
Improving coding-agent scaffolds and benchmarks →
Boosted an internal coding-agent scaffold by >15% on SWE-Bench Verified. Diagnosed harness and environment defects on FeatBench and lifted gold-patch resolution from 37.3% to 99.8%. Released FeatBench-Verified v1.3 and ported parity fixes upstream.
M.A.Sc. Queen’s · ACL 2026, ICML 2026
Research on interpretable & safer LLMs →
First-author paper at ACL 2026 (CoDial: aligning LLM behaviour with explicit dialogue flows). Co-author on RedDebate (ICML 2026, multi-agent red-teaming) and MisLC (Findings of EMNLP 2024, legally-grounded misinformation detection).
Open source
Contributing to LLM agent tooling →
Authored the FeatBench adapter and the trae-agent integration for Harbor (Terminal-Bench’s official harness). Four merged PRs to NVIDIA’s NeMo Guardrails covering parser fixes, prompt rendering, and developer tooling.