Work

A local multi-agent orchestration system. Agents coordinate through an explicit, inspectable loop (plan, act, review) rather than one opaque prompt.

Calendar in, honest week out. AI categorization, a capacity Sankey across five tabs, vibes plus a public booking link, and the same service layer exposed to Claude over MCP.

No single identity can both write code and merge it. Edit agents only propose; branch protection and a separate merge actor decide what ships.

A pipeline that reads architecture diagrams and emits Terraform that already satisfies org guardrails, opened as a reviewable pull request.

You steer. The right specialist surfaces. Drive Mode is the open-source, local-first product family at drivemode.ai: Cursor Drive, Claude Code Drive, and Cline Drive (cline-drivecode).

A toolkit for measuring how fragile or resilient hidden-information schemes are (LSB, invisible Unicode, visible watermarks) across text and images.

An open standard for how AI agents discover, read, and hand off project context, plus schemas and tooling that make it real.

A CLI that catches backend/frontend contract drift in CI (split repos or monorepo) and returns a machine-readable verdict.

A friend and I built this Godot 4 party game in one night: a signal-driven architecture with a headless test harness that runs a thousand shuffle iterations and a hundred random matches to check the rules hold.

A hybrid ML and LLM pipeline for inventory classification, evaluated on held-out data. Research presented at AIM 2024.

Every model interaction takes one governed path: access brokered centrally, inputs checked, outputs logged, so operators can see and control how models are used.

An evaluation SDK that turns LLM interactions into structured cases, scored with independent metrics so agent behavior is testable.

Compiled Mermaid and SVG diagrams with regression checks, so architecture pictures update when the system changes instead of rotting in a slide.

Custom checks for AWS limits that native tooling does not monitor (security groups per VPC, access keys per user), surfaced as metrics that warn before growth becomes an outage.

Read-only evidence gathered into a destruction plan for orphaned cloud resources; a human operator still performs the delete.

An audit that finds where sensitive data lives (file, column, match count) and reports locations without emitting the values themselves.

Four CLIs (autofix, dispatch, draft-flip, merge) share one rate-limited GraphQL client so BugBot threads get fixed and gated without silent auto-merge.

Research-ops scaffolding: ten ADRs, gated codegen rules, and a CI harness before the newsletter triage pipeline itself is implemented.

A Cursor plugin that runs OAuth diligence (cited ToS and pricing research, registration checklist, verification plan) before any integration code is written.

A multi-root workspace that coordinates a Python backend, a React frontend, and Terraform infra for a job-search product, without merging the repos into one monorepo.

Positioning and a minimal landing surface for a product that is still early: clear claims, no fake shipped status.

Spire treats Claude Code config like a Slay the Spire run. It scans a repo, deals a class-specific starter deck of skills and rules, then grows that deck as you clear work.

A personal quant stack: research notebooks, a commission-aware backtester, a dashboard, and infra to run experiments. Research-grade, not a marketed trading product.

A gated pre-publish CLI: inventory, scan, stage, review, and approve with a decision ledger so secrets and PII do not reach a public remote.

A nine-section agent context format with a six-pass compression pipeline, MCP and CLI prototypes, and a roadmap that stays honest about what is not built.

A local-first pipeline that gathers commits, IDE and CLI sessions, ticket references, and notes into structured activity a developer can review.

A skill that makes diagrams the default agent artifact whenever the content has order, connection, state, or proportion.

A skill that turns repo evidence into architecture diagrams by default, with preview-safe HTML and split sources so large sets stay readable.

A tiered, evidence-backed connection graph and multi-page hub preview, so you can see how work links before you write the hub copy.