Building practical systems that make advanced AI easier to understand, use, and ship.
I’m an MSc student at the University of Alberta exploring the space between AI research and useful software. I like taking difficult ideas—attention mechanisms, distillation, agents, learning workflows—and turning them into implementations, tools, and explainable experiments.
My current work clusters around three questions:
- How can language models learn more efficiently?
- How can agents use tools and evidence more reliably?
- How can technical knowledge become a better learning and creative workflow?
| Project | What it explores |
|---|---|
| DeltaAttention | A readable PyTorch implementation of Kimi Delta Attention with a small hybrid language model. |
| on-policy-distillation | Student-visited-sequence knowledge distillation for causal language models. |
| Project | What it explores |
|---|---|
| tradeAgents | Evidence-linked, multi-agent research briefs for US equities—decision support, not automatic trading. |
| LearningAgent | A native Pi learning harness that keeps lessons, side questions, checks, and notes structured. |
| Project | What it enables |
|---|---|
| ai-3d-asset-pack-pipeline | Turns generated concepts into coherent, validated, and packageable Three.js asset packs. |
| ai-podcast | Evidence-backed Chinese AI podcasts with research, balanced scripts, stable voices, and delivery artifacts. |
| xiaohongshu-ai-carousel | Mechanism-first Chinese AI learning carousels with source evidence, image generation, and QA. |
research → build a small working system → measure the failure modes
→ explain the mechanism → package the workflow for reuse
I value readable implementations, explicit assumptions, evidence-linked outputs, and tools that remain useful after the first demo.
Topics: attention and state-space models · knowledge distillation · RL and test-time compute · multi-agent systems · retrieval and research tools · local inference · technical education
- GitHub: @Playitcooool
- X: @Liemar10539579
- University of Alberta: ualberta.ca
Thanks for stopping by. I’m always interested in thoughtful conversations about AI systems, research, and tools.