Explore the systems behind this map ↓
Explore commits · Review pull requests · Browse repositories
How the live clock works
The days, hours, minutes, and starting seconds are calculated from GitHub's public account-creation timestamp when the asset is generated. The seconds readout and hand then advance once per second while the SVG is open. GitHub caches README images and does not run custom JavaScript, so the daily snapshot is the authoritative value; the moving seconds are a live visual continuation, not a wall-clock service.
I study how intelligent systems are built, measured, and made trustworthy. My work connects agentic AI, benchmark production, evaluator integrity, embodied intelligence, and research infrastructure.
The through-line is simple: turn ambiguous technical questions into evidence that can be inspected, challenged, and used.
| System | What it demonstrates |
|---|---|
| Agents' Last Exam | An evidence-led architecture for producing, evaluating, and governing private agent benchmarks. |
| Research Portfolio | Deep research on vision-language-action models, world models, and emerging embodied-AI systems. |
| BaZi Context Agent | A local-first prototype that separates deterministic reasoning, user-controlled context, and AI-assisted interpretation. |
| Info Collector | A research workflow that turns public reading into structured, traceable, reusable knowledge. |
- How do we evaluate configured agent systems rather than isolated models?
- What makes an automated evaluator valid, auditable, and resistant to gaming?
- How can research operations preserve evidence without slowing down iteration?
- Where do world models, VLA systems, and agentic workflows meaningfully converge?
Project commits and pull requests use GitHub's author-search surface across the four repositories listed above. Code proportions use GitHub Linguist bytes across the same scope. The research orbit maps visible work; it is not a performance score.