Simon Heikkila

Short — 30 seconds

Who: ex-veterinary surgeon, ex-ASX-listed-company product manager, now a one-person product team running an AI agent fleet from Brisbane.

Proof: rescued a stalled outsourced website rebuild for one of Australia's top-rated mortgage brokerages — took over the codebase, finished it solo, zero-downtime cutover, then shipped a quarter's roadmap in the three weeks after launch. Conversions up, measured to a standard that survives significance testing.

The honest number: 81,000+ GitHub contributions in the last year across 66 repositories — fewer than one in five shipped. The ratio is the credential: enormous output, brutal selection.

Wants: complex problems in regulated industries, with people who move fast and value the work. Contract or fractional preferred.

That's the 30 seconds. The two-minute version has the numbers.

Medium — 2 minutes

I design and build production software at the speed of an AI-native workflow — delivering in weeks what conventional teams schedule across quarters — mostly for regulated Australian property and financial services, where the interface has to convert and survive compliance review.

Selected outcomes

• Rescued a stalled outsourced site rebuild (WordPress → Next.js) for a high-traffic financial-services business: took ownership of the codebase, finished what the vendor couldn't, ran a zero-downtime staged cutover — then shipped a quarter's roadmap in three weeks post-launch.
• Post-launch conversions up, measured to a standard that survives significance testing — replaced the vendor's underpowered A/B reporting with honest measurement.
• Lifecycle estate that held conversion steady at 14× volume: 200+ HubSpot workflows; pre-approval→settlement held at 86–89% while annual volume grew 84 → 1,200+. Weekly finance newsletter since 2019: 370k+ emails at 30.9% open; its dormant pool supplied 14–46% of settlements.
• $32M in closed fundraising supported by pitch decks and financial models — four confirmed-closed syndication deals plus a $5.6M oversubscribed fund.

Experience

Capabilities

UI/product: design systems, complex form/workflow UX, calculators, conversion optimisation, A/B testing · AI-native engineering: multi-agent orchestration (Claude/Codex), Temporal durable workflows, Next.js/React/TypeScript, edge workers, Astro, HubSpot · Regulated delivery: AU financial-services compliance UX, disclosure copy, audit trails, security hardening · Product leadership: roadmaps, stakeholders, pitch/investor narrative

Long — the whole story

Everything above, plus the parts a PDF never has room for.

The unusual start

Seven years as a veterinary surgeon in Australia and the UK — small-animal surgery, emergency work, sole-charge shifts where you diagnose under pressure with incomplete information and no one to hand off to. As a locum I grossed 30% above practice average. Everything since is the same job with different patients: find the root cause, not the symptom.

In 2015 I wrote an email sales funnel for a friend with an audience he'd never monetised. It made $60k in two months. That got me hired at ChildsPass, a kids' activities marketplace in Vietnam — where I complained about the product until they invited me to build it instead. Team of four to eight; partner locations zero to 600 in six months.

The two-weeks story

At Simble I said their development system could run better. The incumbent product lead disagreed — strongly enough to bet I couldn't prove it. I asked for two weeks. Two weeks later the system was singing — and I didn't take the role so much as absorb it, because the processes I'd built were self-sustaining underneath me. I ran a LeSS framework across three teams of eight developers, managed a full product pivot, and as Operations Manager ran Finance, People & Culture, and Risk & Compliance for Australasia through the run-up to the ASX listing. I left a few months before the IPO.

The arc — how I got AI-native

Twelve months ago I started what looked like a Christmas toy: a 2D falling-sand game. It became a semi-functional DEM/FLIP particle simulation at sluice-and-environment scale, GPU compute in Rust — from someone who had never written a line of shader code. The three dead sim repos behind it are what learning at that speed costs.

Then I ran my work through other people's agent infrastructure — Gastown, then OpenClaw — to learn the patterns worth stealing and the walls worth avoiding. Then I built my own, twice: chum and chum-v2, back when traces as a meta-layer of reinforcement was just becoming a conversation. It turned out you need more than one hubristic vet-turned-product-manager to make that market happen — but along the way I learned to bootstrap a Temporal-based, always-on agent team on a VPS, with durable workflows and retry semantics that respect always-on work.

The fleet I run today — Claude Code as orchestrator, subagents scoped to one job each, parallel git worktrees, an always-on VPS with 13 scheduled jobs across 46 projects — runs on what chum taught me. I'm model-agnostic by habit — I've been riding this curve since GPT-2, and today I work daily across Claude, GPT and GLM, with detours through Gemini and most of the rest. The honest yield: 66 repositories started, under 20% shipped, 81,000+ contributions in the last year. Verification hooks in my build process block any "done" claim that lacks an attached artifact. Failure is part of the process; the half-finished repos are the exploration budget, and the finished ones are what the budget bought.

The flagship — a rescue, end to end

One of Australia's top-rated mortgage brokerages had a specialist outsourced vendor rebuilding their WordPress site in Next.js. It stalled. I took ownership of the codebase, finished the design system, calculators and intake flows, and ran an edge-worker staged cutover to production with zero downtime — Googlebot handled explicitly, form and API paths pinned, rollback runbook ready and never used.

In the three weeks after launch I shipped what's usually a quarter's roadmap: a refreshed calculator suite, overhauled lead-capture flows, FY2026-27 regulatory updates across every state scheme, security hardening, SEO. Conversions are up — and measured to a standard that survives significance testing, because I replaced the vendor's underpowered A/B reporting with honest measurement. I'd rather report a defensible "up" than an impressive number that dissolves under scrutiny.

The same engagement includes the lifecycle work: a 200+ workflow HubSpot estate, from pre-approval expiry ladders (54–59% open rates) to a three-year post-settlement nurture, that held pre-approval→settlement conversion at 86–89% while annual volume grew fourteen-fold. A weekly finance newsletter running since 2019 — 370k+ emails at a 30.9% open rate — maintains the dormant pool from which 14–46% of settlements are later drawn. The longest revival settled after 750 days dormant.

The other lane

$32M in closed fundraising supported by pitch decks and financial models I built — four confirmed-closed syndication deals plus a fund that closed oversubscribed at $5.6M. Narrative is a product surface too.

What I want

Complex problems, in regulated industries, with people who move fast and value the work. Contract and fractional engagements preferred — fixed-scope sprints, production builds and rescues, or an ongoing product partnership with a defined mission and a standing exit.