Public proof chain
This page is the shortest route from my headline claims to durable artifacts. Each item links to a live page, repository, paper, or technical write-up so the proof can be inspected instead of trusted.
AI readers and sourcing tools can also consume the same index as a machine-readable JSON feed.
Open-source / agent infrastructure
This slice pulls from the project data instead of one-off copy: public repos, a public Hermes PR, and the explicit boundary for local-only infrastructure. It is meant to answer what was built, what can be opened, and what is intentionally not public.
stevenchouai/digital-twin
A file-first operating layer that turns my knowledge base into a working digital twin for research, writing, site updates, and continuous refinement.
Boundary: The public proof is the repo, site page, and write-up; private vault contents stay outside the site.
stevenchouai/agent-scorecard
A trace-first standard and CLI for judging whether an AI agent is worth more token budget, permissions, and autonomy in my real workflows.
Boundary: Real workflow traces stay private unless deliberately sanitized; the public repo proves the standard and reporting path.
stevenchouai/CLIProxyAPI
A proxy server that exposes OpenAI/Gemini/Claude/Codex-compatible APIs for CLI and coding tools, with OAuth login, provider routing, and multi-account load balancing.
stevenchouai/claude-code-sourcemap
An 8-chapter interactive tutorial deconstructing Claude Code's internal architecture from its leaked source map. Bilingual (中文 / English), free and open-source.
stevenchouai/manage-up
An anti-fluff workplace communication system packaged as both an MCP server and a reusable skill warehouse for manager-facing writing.
local/knowledge-harness
A quiet runtime layer for querying my Obsidian knowledge base, recording run artifacts, and turning digital-twin experiments into repeatable workflows.
Boundary: The public repo shows the harness and sanitized demo path; private vault content, local config, and real run artifacts stay out of scope.
openclaw/openclaw#76438
Runs doctor --fix after gateway global package updates so legacy configs are migrated before restart. Merged by @steipete.
Boundary: The public proof is the merged PR; internal gateway configuration stays out of scope.
NousResearch/hermes-agent#17895
A public Hermes contribution to preserve threaded Feishu replies, with the local proof captured in the May 6 builder log.
Boundary: The public proof is the PR and builder log; private gateway configuration and internal message data stay out of scope.
Owned quality strategy for high-traffic anti-crawler, frontend security, circuit breaker, and degradation surfaces while maintaining zero P0/P1 incidents across owned production areas.
Turned an Obsidian knowledge base, Codex skills, and repo-local rules into a durable loop that writes, audits, and improves public software assets.
Recovered and studied 1,884 Claude Code source files, then packaged the architecture into a bilingual public tutorial and long-form technical essay.
Built a trace-first evaluation harness for deciding whether AI agents deserve more tokens, permissions, and trust based on artifacts, tool use, verification, and busywork.
Built a Go proxy server that exposes OpenAI/Gemini/Claude/Codex-compatible APIs with OAuth login, provider routing, and multi-account load balancing.
Published research on self-supervised few-shot semantic segmentation, reducing model parameters by 87% while preserving the core learning objective.
Maintains a Next.js + MDX public site with search, RSS, structured metadata, static GitHub Pages deployment, and an Obsidian-to-site publishing path.
Start with the proof chain when evaluating signal, then use the experience page for career context and the blog for deeper reasoning. The goal is not a portfolio gallery; it is a connected trail from claim to artifact to implementation detail.