Launching this portfolio
Finishing the Astro 5 + Axum 0.8 monorepo — WASM Black-Scholes demo live, case studies approved, now waiting on DNS/CF Pages decisions before the production flip.
Now
What I'm focused on, May 2026
Building this portfolio, running an autonomous trading system, and deepening multi-agent orchestration patterns.
Autonomous Kalshi trading system
Running a halt-safe paper-mode strategy fleet with atomic kill-switch state machines, SHA-sidecar verification, and a six-wave swarm that closed 13 audit P0s without a single halt-state drift.
Multi-agent CLI fleet orchestration
Designing and iterating on agent swarm patterns — parallel dispatch, orphan-runner detection, concurrent edit-race avoidance, and halt-contract enforcement across Claude Code CLI and Codex CLI.
Rust as the default for anything compute-critical
The WASM pricer is the most visible example, but the same preference drives API services, CLI tooling, and any path where I want type-system guarantees at integration boundaries.
Reading deeply on distributed systems and production AI
Working through DDIA, async Rust, and recent multi-agent orchestration papers — less to collect references and more to sharpen the mental models I reach for when something breaks at 3am.
Churn-resistant tooling decisions
The 2026-04-28 dev-tooling rollout crystallised a principle: prefer terminal-native, prefer composable, prefer tools whose failure mode is loud rather than silent. Carrying that into every new addition.
This page is a snapshot of what I’m actually working on right now — not a résumé, not a highlights reel. The idea comes from Derek Sivers’s /now movement: a single public answer to “what are you up to these days?” that stays honest because it’s updated rather than curated.
Building the portfolio. The stack is Astro 5 for the frontend, Axum 0.8 for the API, and a Rust WASM island for the Black-Scholes options pricer demo. Most of the engineering work is done — the current blocker is a set of operator decisions: Cloudflare Pages project name, API hosting path, final domain. Once those resolve, the deploy is a single pipeline run.
Running the trading system. The Kalshi strategy fleet has been in paper mode since April. The focus lately has been correctness: halt-state atomicity, SHA sidecar verification across multi-agent swarms, and closing orphan-runner bugs where a safety mechanism was hardened at the helper layer but the cron runner supplying its inputs was still passing hardcoded constants. Real live arming is gated on wall-clock evidence accumulation — not code fixes.
Sharpening the agent patterns. Three months of agent swarm work has surfaced a consistent class of silent failures: field-selection drift between writers and readers, completion notifications that fire while an agent is still issuing tool calls, and safety gates that are structurally correct but functionally inert because the runner never passes real inputs. Writing these up as reusable skills so the next swarm doesn’t rediscover the same class.