Twelve systems, built solo. One, the Octoport, is the business. The rest are the proof.
An intelligence engine. Bring any question you can describe in language. The engine returns sourced, signed intelligence shaped to what you actually asked: what's known, what isn't, what the evidence supports, what to do next. Same engine, whether you're reading a market window, a research position, a career pivot, a technical bet, or a decision that does not fit a category. Domain-agnostic by design.
Octothorped stands on a foundation: a deliberate run of systems for making AI trustworthy. Open code you can click, private builds you can see, and the design concepts in between.
Every project here is a public repository. Click any to read what it does and see the real code. Nothing is mocked or inflated.
A local-first creative-writing environment: a local model with RAG over your own canon, so the story stays consistent and you can talk to your characters in their own voice. Read the full chapter below.
→ ran in 2025 · a paused passion project · the chapter ↓
An AI video-production suite that turns written course modules into finished video, with a human-in-the-loop script editor and a smart regeneration system that adapts to your edits. React front end, Python back end.
→ full-stack app · built 2025
You describe a vision; "The Director" converses, plans a Scene → Transition → Scene structure, and outputs a single structured production blueprint — shot list, a real camera-command grammar, music — engineered to drive a video-generation API. A fully specified agentic system.
→ design spec, not a shipped app
Not a chatbot — a private AI built as a cognitive mirror for an ADHD mind: designed to collapse the distance between an idea and its execution to near zero, and to translate non-linear, high-velocity thought into structured, actionable output. A Socratic partner mandated to question premises and surface gaps, not a tool that simply obeys.
→ a full system constitution + interface design
It started as Weaver, my first AI project. A rough agent built before I knew what my hardware could carry, with an over-ambitious pipeline that taught me, mostly by failing, exactly what to ask of a machine and what to stop asking. Weaver's Sanctuary is what it became once I'd learned.
I wanted a new way to interact with a story. To write it, and to understand it, from the inside. To get into the heads of the characters and find out why they made the choices they made, grounded in the story itself. The systems thinker in me took the act of writing apart from two angles: the author looking in, and the story understood from within. What came out is a way to connect a writer to the gravity of their own words. Each word matters. Each change ripples. A whole plot hole can open from a single typo.
Underneath, it's a full-stack application: a local model, a vector store, retrieval over your own canon. But the plumbing isn't the point. The point is that the story remembers itself. A detail written three chapters ago doesn't quietly vanish. You can sit down and talk to someone from your own story, in their voice, and they answer in character. Not a chatbot. The person you wrote.
I'm a writer, and I share my life with an avid reader. This is the tool I built out of love for the thing she loves, and a quiet belief that it could change how an author looks at the page. I won't insist on that. The work is what it is.
Weaver's Sanctuary isn't open source. It ran in 2025, and I haven't touched it since; it's a paused chapter, not a product. It's here because it's true, not because it's finished. The repos above are proof I build for machines. This is proof I build for people.
Octothorped runs on your market. Or the foundation above is the proof I can build the system you need — agentic systems, RAG, verification infrastructure, from first principles.