Skip to content

—Work in progress

An assistant that runs in your browser, not on my server.

I build agentic and RAG systems for a living, so this site should eventually be one. The plan is a local-first assistant you can interrogate about my work — running entirely on your device via WebGPU, grounded in structured data so it cannot make things up.

WebGPU · checking

Checking your browser…

No API key, no server, nothing sent anywhere. That is the whole point.

  1. 01Done — shipping in the agentic-profile repo

    Grounded profile store

    The same structured JSON memory that drives my agentic-profile project, so answers are retrieved rather than generated from a model’s vague recollection of a resume.

  2. 02Next

    In-browser inference via WebGPU

    WebLLM loads a quantised model into your GPU and runs it locally. No API key, no server, and nothing you type leaves your machine. Once cached, it works offline.

  3. 03Designed

    Tool-calling agent loop

    Deterministic JSON tool calls against the profile store, project index and Credly data, with an explicit refusal path when the answer genuinely is not there.

  4. 04Exploring

    Agentic workflows

    Longer-running tasks — read a job description, compare it against my actual experience, and produce an honest gap analysis rather than a sales pitch.

Until this is live, the working version runs locally through Ollama. See the agentic profile repo ↗ — or just read the projects the old-fashioned way.