Toran Bruce Richards Portfolio
Creator of Auto-GPT (187,000+ GitHub stars, one of the fastest-growing open-source repositories in GitHub's history), one of the first widely used experiments in fully autonomous, self-prompting agent loops that break a high-level goal into subtasks without a human in the loop at each step.
Location
Didn't say. We're respecting the mystery.
Skills
Notable
- Auto-GPT: 187,000+ GitHub stars (one of fastest-growing repos in GitHub history)
What makes it work
A breakdown of the choices that make this portfolio stand out.
An early, extreme version of an architecture the field then had to rein in
Auto-GPT's fully autonomous, no-human-in-the-loop design demonstrated both the appeal and the real failure modes (infinite loops, runaway API costs, hallucinated subtasks) of unconstrained agent autonomy — the project's limitations directly shaped the more constrained orchestration patterns that followed.
Scale of adoption as a historical marker, not just a vanity metric
Growing to one of GitHub's most-starred repositories within weeks of release marks a specific moment when public interest in autonomous agents went mainstream, itself a reference point other agent-architecture writing cites.
A named, standalone project rather than a fork or wrapper
Auto-GPT was a from-scratch architecture, not a tutorial fork or wrapper around someone else's framework, giving it its own design decisions to study rather than inheriting them.
What AI Agent Architects can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
If you ship an early, ambitious version of a new architecture, expect its failure modes to become part of its legacy too — document them, since the limits of a first attempt often shape the field's later, more careful designs.
A from-scratch project, even a rough one, is more instructive to study than a wrapper around an existing framework — build your own architecture at least once.
Recognize when your project's timing is itself the differentiator, not just its final technical sophistication.
Projects
One of the most-starred repos on GitHub, pinned next to two small utilities built on the same idea: describe a job to an LLM and let it do the work. Together they show that idea at three scales.
It started as an experiment in self-prompting agents that break a goal into steps and run them. It is now an open-source platform: you describe an outcome in plain English or build it block by block in a visual builder, then run agents on demand, on a schedule or from a trigger.
It took a viral experiment and kept it going as a product, adding a hosted platform, a marketplace and a builder, which requires more than one good idea.
Four surfaces: AutoPilot, Agents, Marketplace, Build
Hosted platform or self-host, with agents run on demand, on a schedule or by trigger
187.6k stars and 46k forks
A small library where you pass a function signature, arguments and a plain-English description, and GPT-4 returns the output as if the function had run. It needs no implementation. The README credits Ask Marvin as the inspiration.
It shows early thinking about LLMs as callable components with a function contract, an idea now built into function calling and structured outputs.
Signature + args + description → model-generated return value
One call: ai_function(function, args, description, model)
Model is swappable, with GPT-4 as the default
A set of LLM-powered scripts for maintaining a GitHub repo. They find duplicate issues by similarity threshold, draft and post GPT-4 replies to issues, review pull requests, and generate architecture overviews and READMEs.
Maintainer tooling built by someone running one of GitHub's busiest repos, aimed at real triage work.
Duplicate-issue finder with tunable threshold and time window
GPT-4 issue responder and pull-request reviewer
Split read-only vs read/write tokens
More AI Agent Architect portfolios
Ready to build your portfolio?
Follow the AI Agent Architect roadmap — skills, projects, and timeline to get hired.