Personal projects · Open source

Featured projects

Selected personal projects, described from their reviewed READMEs.

  1. Personal project

    Local Crew

    A local-first orchestration harness for multi-device AI inference networks.

    Local Crew pools Ollama, OpenAI-compatible and Anthropic endpoints across a local network into one managed set of inference nodes. Named agent identities stand in for the hardware, and a delegation layer routes work by capability, availability and memory headroom.

    • An autonomous mode that fills a task queue through three-phase consensus, routes and verifies the work, and recovers from failures
    • Persistent memory: directives, roadmap, changelog and each agent's private memory carry across sessions
    • Built for the local network: the server rejects non-local clients by default
    • A command line, an API and a browser interface

    AI workflow + AI product reliability + Full-stack delivery

    TypeScript · MIT licence

  2. Personal project

    Orgos

    An agent-first organizational operating system.

    An observable local environment where shared infrastructure, organization data, staffed agent identities, flow sequencing, memory, files and text assets all live on disk in readable JSON, Markdown and plain text, run from a desktop-style shell of applications.

    • Flow sequences objectives continuously instead of on a human calendar
    • Queue dispatches inference and shows node telemetry
    • Meet, Mail and Staff hold group discussion, agent and human messaging, and staffed identities
    • A Bun server with unit, integration and regression tests

    AI workflow + Context & team rules + Full-stack delivery

    JavaScript · Bun · MIT licence

  3. Personal project

    Promptpad

    A local-first prompt drafting tool.

    Promptpad turns terse instructions into polished, copy-ready prompts with a local model through Ollama: refine a brief into a detailed prompt, edit it, then reinforce it for production.

    • A low-temperature cleanup pass removes meta framing such as “Here’s…” when a model adds it, keeping output clean across tiny and large local models
    • Real-time token counting and a live preview while editing
    • A command-line interface alongside the web app

    AI workflow + Context & team rules + Interface & spatial

    TypeScript · Next.js · MIT licence

  4. Personal project

    Synthetic Data

    A browser-only synthetic dataset generator.

    Draft a JSON schema, connect a local Ollama server and generate small synthetic datasets without sending anything to third-party services. There is no build step: open the page and start.

    • An inline schema builder with field controls, a record count and a live preview
    • A connection panel that lists the local server's models
    • A generation dashboard with phase and status counters, an event log and the dataset as a table
    • Its README is candid that error handling is basic: failures show in the event log and records are retried

    Test architecture + AI workflow + Interface & spatial

    JavaScript · MIT licence

  5. Personal project

    Stay Ahead

    A minimal template for an agent-driven workflow.

    A small JavaScript codebase with unit tests, UI components and a service layer, set up to show how several agents can collaborate through test-driven development and code review.

    • One quality gate in the build: lint, type-check, unit tests with Jest and end-to-end tests with Playwright
    • Agent roles: Claude for high-level features and commentary, Codex for code and tests by TDD, Copilot for inline suggestions
    • Agents plan before coding and keep the documentation in step

    Test architecture + AI workflow + Collaboration

    JavaScript · Jest · Playwright · MIT licence

Project descriptions follow the reviewed READMEs.