engineer. builder. systems+ai.
I work full-stack on payments, bridging, and staking at Polygon, focused on shipping systems that improve speed and confidence through testing, reliability, and clear docs. Outside work, I build with AI and use personal projects to sharpen agent orchestration in real-world workflows.
Last updated: 2026-05-18
The active surface area: production work, public tools, and operator systems.
Polygon's native liquid staking product for staking, unstaking, migration, and DeFi-ready sPOL positions.
Add more transport modes and explore notification workflows.
Keep dogfooding the operator surfaces and tighten what is genuinely useful versus just interesting once.
A few receipts before the text starts doing too much work.
agent-first public data tools
A family of terminal tools for public data: transport, companies, rail, fuel, and Parliament. Built for humans, predictable enough for agents.
$ tfl status Central: Good Service District: Minor Delays $ parliament search bills "energy" --take 3 Great British Energy Bill 5172 Committee
proof: five published CLIs · npm packages · landing pages · reusable agent skills
operator tooling
Local-first web dashboard for Hermes Agent. Runtime health, sessions, cron, skills, memory, files, and usage in one place.
proof: real local agent state, not fake dashboard filler
full-stack AI product
Roast and glaze photo app with image uploads, auth, credits, and share-friendly output loops.
proof: upload → generate → credits/payment → share loop
Current focus across work, public tooling, and the personal systems behind both.
At work: stepping into Polygon's Open Money Stack while keeping the bar high on reliability, testing, docs, and observability.
Outside work: building agent-first CLI tools that are actually useful; tfl-cli, companies-house-cli, rail-cli, fuel-cli, and parliament-cli are the clearest proof right now.
Personal systems: tightening family planning, nutrition, and fitness so the setup supports the rest of the work instead of competing with it.
07 May 2026 · 6 min read
AI has made code generation cheap, and everyone is all in, but nobody is thinking about how the review cycle should change.
27 Apr 2026 · 6 min read
Vibe coding is one way to use AI. Treating every AI-assisted workflow as the same thing is lazy.
17 Apr 2026 · 3 min read
Agents can turn a day into a week of code output. That still doesn't mean the work was the right bet.
Tightened the Hermes-native operating contract with clearer modes, accountability loops, and memory hygiene.
Published new agent workflow skills to the public shan-skills repo.
Shipped a backend gateway PR with typed client generation, a normalized endpoint, OpenAPI spec, and caching.
Raised a support triage doc and landed e2e testing improvements to reduce non-reproducible app issues.
Hardened personal projects by moving them to pnpm and adding a 3-day minimum package release age.