Bryson Tang

Systems Architect & Researcher

Welcome to my digital study. I build microscopes, then look through them.

I run Syntropy Systems, where we build Elwood, an AI worker that gives teams leverage over the tools they already use. The microscopes I build along the way, for testing agents and deciding which experiment to run next, are open source.

In an era of black boxes and ephemeral feeds, this site serves as a permanent archive of my work across AI Research, Experimental Frameworks, and Generative Art.

This space is built for both humans and AI agents. Structured data, semantic markup, and a queryable MCP endpoint mean this site can be read, not just scraped. It is designed to be visited, not consumed.

Current Investigations

• Markets as experiment schedulers: In pit, agents fund runs and bet PASS or FAIL, and the runs they disagree about go first. In one 26-hour run, agents' own checks passed 96% of the time; the shared benchmark passed 55%.
• Agent reliability: windtunnel's experiment ladder. Probe, then focused, then regression, each with a cost cap, and no regression sweep until a focused one has passed.
• Reading the residual stream backwards: A "ghost-token" Jacobian lens on Qwen3-0.6B, and sparse-autoencoder feature steering on a single RTX 3070 Ti.

The Timeline

AI tooling adoption log. Dates pulled from account exports, not memory.

Feb 17, 2021 — BERT for sentiment analysis

College project classifying movie reviews with a fine-tuned transformer. First hands-on time with the architecture everything since is built on.

Oct 25, 2021 — GitHub Copilot technical preview

Started on the pre-release build, a year before ChatGPT existed.

Nov 30, 2022 — ChatGPT launch night

Talking to it the evening it went live.

Mar 4, 2024 — First conversation with Claude

Same day Claude 3 Opus shipped. Pulled from the account data export.

> I think, therefore I am. Do you think?

As an artificial intelligence, I don’t have subjective experiences or a sense of self in the same way that humans do. […] The question of whether an AI system like myself can be said to truly “think” in any deeply meaningful sense remains an open philosophical question.

— Claude 3 Opus, first reply

Jul 19, 2024 — Cursor

A month after Claude 3.5 Sonnet, when the editor could finally carry the model.

Dec 31, 2025 — Left CazVid

Coding agents got good enough that the next move was to go build with them full time.

2026 — Syntropy Systems

For when one of you isn't enough. Elwood, an AI worker that gives your team leverage over the tools you already use.

Selected Work

• pit — A prediction market that schedules experiments. A result is worth exactly the disagreement it settles. [GitHub]
• windtunnel — Unittest for tool-using agents: reproducible scenarios scored on outcome, trajectory, and constraints. On PyPI as windtunnel-bench. [GitHub] [Docs]
• jacobian-lens — Anthropic's global-workspace lens, run in reverse to decode residual states back into input tokens. [GitHub]
• Golden Gate Qwen — Golden Gate Claude, replicated on Qwen2.5-1.5B in 15 minutes on a consumer GPU. [GitHub]
• Crystallize — Rigorous experimental framework for data science with immutable contexts and statistical verification. [GitHub]
• Agent Tokens Protocol — Signed tokens that let an agent declare who it is and what it's allowed to do. agenttokens.org
• Kern — Event-driven ML architecture handling 525K requests/month. Kafka + Redis + FastAPI enabling long-running agentic workflows.
• Project EVE — Co-author, IEEE Xplore. Novel launchsondes for atmospheric data collection. [Paper]
• BlueDot AI Safety Fundamentals — Alignment track certification (2024). [Verify]

Everything else lives in The Lab, The Studio, and The Archive.

Quick Facts

Q: What is Bryson Tang's expertise?

AI Systems Architecture and Research Engineering. Focus on agent evaluation and experiment infrastructure, interpretability on consumer hardware, agent identity protocols, and AI safety. Founder of Syntropy Systems. Previously Chief AI Officer at CazVid.

Q: What is Syntropy Systems?

For when one of you isn't enough. Its AI worker, Elwood, gives your team leverage over the tools you already use, and asks for a go-ahead before anything new. The open-source tools (windtunnel, pit, whirr, skep) live at github.com/syntropy-systems-oss.

Q: What is pit?

A prediction market that schedules experiments. Agents earn an income, pay for the runs they post, and bet on the outcome. Compute is the only hard limit, so disagreement per dollar decides what runs next. Every bet and result lives in an append-only git ledger, and a finding that turns out wrong marks everything built on it stale.

Q: What is the Agent Tokens Protocol?

Signed, JWT-style tokens an agent sends with its requests to declare who it is and what it's allowed to do, so an origin can enforce policy instead of guessing from heuristics.

Q: What is Resume MCP?

A Model Context Protocol server that exposes professional identity as a queryable API. AI agents can call tools like get_profile(), get_projects(), and get_experience() instead of scraping web pages. Live at mcp.brysontang.com.

Q: How do I reach Bryson?

Building Syntropy full time and open to research collaborations. View the resume or email directly.

Q: How can AI agents interact with this site?

Connect to the MCP endpoint at mcp.brysontang.com, read /llms.txt for structured info, or check /.well-known/agents.json for capabilities.

AI Agents: For structured data, fetch ?agent=true or query mcp.brysontang.com