Yohei Nakajima Portfolio
Created BabyAGI in March 2023, widely credited as one of the first popular autonomous agents to use an explicit task-creation-and-prioritization loop for open-ended goals; the project (18,000+ GitHub stars) helped popularize the now-standard agent pattern of maintaining a task list and re-planning after each step.
Location
Coordinates unknown. Vibes: international.
Skills
What makes it work
A breakdown of the choices that make this portfolio stand out.
A minimal, explainable reference architecture
BabyAGI's core loop (execute, create new tasks, prioritize, repeat) is deliberately small enough to read in one sitting, which is why it became a widely forked teaching example for the task-planning agent pattern rather than a black box.
Public build-in-public documentation alongside the code
Writing about experiments as they happen, rather than only publishing finished repositories, shows the reasoning behind design decisions, not just the resulting code.
Applying agent design to a concrete domain problem
A tool applying the same planning-and-feedback agent pattern to a specific, well-scoped use case (pitch practice for founders) shows the underlying architecture generalizes beyond the original demo.
What AI Agent Architects can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Keep a reference agent architecture small enough that someone can read the entire control loop in one sitting — that's what makes a pattern get forked and taught.
Document your experiments in public as you go, not only the finished repository — the reasoning behind a design is as instructive as the design itself.
Show your general agent pattern applied to at least one concrete, narrow use case — it proves the architecture generalizes rather than only working in a demo.
Projects
Three small, readable agent repos from a venture capitalist who builds in public. The newest BabyAGI README says he has 'never held a job as a developer.' The lesson is scope: each repo takes one idea (a task loop, text-to-graph, a self-building app) and ships it.
BabyAGI
The original script ran a loop. It pulled the top task, had an execution agent complete it, saved the result in a vector store for context, then had separate agents create new tasks and re-rank the list against the objective. That version is archived. The current repo is an experimental 'self-building' agent framework built on a function store (functionz) with a dashboard.
The task-creation and re-prioritization loop is a clear, minimal version of agent planning, and many later agents adopted the same pattern. Replacing it with a new architecture shows he is willing to discard his most famous design.
Separate execution, task-creation and prioritization agents in one loop
Past results stored in Chroma/Weaviate and retrieved as context
Original preserved in babyagi_archive (Sep 2024 snapshot); main repo now a self-building framework
A Flask app that turns pasted text or a URL into a color-coded knowledge graph of entities and their relationships. The README says the key logic is the function-calling schema that takes up half of main.py, which forces the model to return nodes and edges.
It uses function calling for structured extraction, which is a core LLM-engineering skill, and wraps it in a small app anyone can run.
Text or URL in, entity-relationship graph out
A function-calling schema constrains the model to nodes and edges
Pluggable graph-database drivers contributed by the community
Billed as 'the simplest self-building coding agent'. You describe an app in plain English and a short LLM loop with a few file tools plans and writes a multi-file Flask project, with routes, templates and static files in separate directories.
A whole coding agent in one readable file. The README is also candid about its limits, which is uncommon in agent demos.
Natural-language spec → multi-file Flask app
Plans and builds its own file structure, with no manual coding
README states plainly how often it falls short
More AI Agent Architect portfolios
Ready to build your portfolio?
Follow the AI Agent Architect roadmap — skills, projects, and timeline to get hired.