01 / BUILD WITH AI
AI can multiply the build.
Versioned instructions turn one specification into isolated, reviewable workstreams—without losing the accountable trace.
Independent AI engineering practice/Finland
Software that earns the word production.
I design, build and operate applied AI systems, commercial APIs, analytics products and forecasting infrastructure—one accountable owner from the first spec to the on-call that follows.
Taking selective engagements
From concept to operated system
0103Molly — Book Intelligence
20GitHub repositories
528ktracked source lines
6.3k+agent & coding sessions
440AI-co-authored commits
Re-measured 10 Aug 2026.
GitHub + 10 local codebases.
01Project archive
The rest of the workshop.
Engineered systems across agents, SaaS, forecasting and data infrastructure. Public detail stays intentionally concise where client work or real capital is involved.
P/01Quant & forecastingIn development
edgelab
Pre-registered research, honestly scored
Python · NumPy · SciPy791 tests
P/02SaaSPrivate beta
Iisi
Property & rental operations for Finnish landlords
Next.js · TypeScript · Supabase77 database tables
P/03AI & agentsPrivate beta
Multi-Agent Dev Room
Real-time collaboration layer for coding agents
TypeScript · React · Node.js4 agent clients supported
P/04Quant & forecastingIn production
Polymarket Trading Engine
Low-latency signal engine for prediction markets
Python · WebSocket · asyncio14ms configured decision cadence
P/05Quant & forecastingIn development
NHL Quant
Ground-up forecasting stack for pro hockey
Python · DuckDB · Dockerin development build phase
P/06Quant & forecastingPrivate system
Futis — Player Ratings
TrueSkill ratings for the Finnish league pyramid
Python · SQLite · TrueSkill Through Time1.16M player appearances
P/07AI & agentsIn development
Selko
A product that is allowed to answer "uncertain"
TypeScript · React · Node.js3 answer states
P/08Data & pipelinesPrivate system
Feed Latency Benchmark
Built to refuse a conclusion, not to flatter one
Python · PostgreSQL · asyncio51 blocking gates
02What I build
Ambition, with the infrastructure underneath.
The polished interface is only the visible layer. Product logic, data integrity, deployment and operations are designed as one system.
01
Applied AI that can be trusted
LLM products, agent workflows and provenance systems with evaluations, visible uncertainty and fail-closed boundaries designed in from day one.
Claude APIOpenAIMCPEvalsAgents
02
Platforms built to carry load
Commercial APIs, metering, auth, billing, real-time data and operational tooling—designed as one coherent system rather than a chain of demos.
TypeScriptPythonGoPostgresRedis
03
Models with evidence attached
Forecasting, market intelligence and research pipelines where leakage, uncertainty and integrity gates are measured instead of hand-waved.
NumPySciPyDuckDBEChartsBacktesting
03How it holds up
AI writes most of the code. The engineering is the harness.
High throughput is useful only when the output remains explainable, reviewable and safe to operate.
Operating principle
Verification is authoritative,
not advisory.
- 01
Frame the consequence
The first deliverable is a precise operating model: users, failure modes, data boundaries and machine-checkable exit gates. The build starts only when ‘done’ can be proven.
▸spec → acceptance gates → measurable release decision - 02
Build inside guardrails
AI agents increase throughput, while version-pinned instructions, isolated worktrees and narrow interfaces keep parallel work reviewable. Speed comes from the harness, not from skipping judgment.
▸440 AI-co-authored commits · 6 tracked instruction files - 03
Try to disprove it
Tests, evaluations, provenance checks and independent review are authoritative. A check that cannot fail is not a check, and a claim that cannot be verified does not ship.
▸2,000+ automated tests · failure paths designed first - 04
Own what happens next
Deployment is the middle of the work. Monitoring, quotas, incident paths, migrations and the on-call feedback loop stay in the same engineering context as the product decision.
▸spec → build → deploy → observe → improve
A serious system deserves one accountable owner
Bring me the system
that has to work.
Product builds, API and data infrastructure, or a senior AI engineering role with genuine production ownership. If the problem is real, let's make the software hold.
Start with an emailcomanjoh@gmail.com↗End-to-end product buildAPI & data infrastructureSenior AI engineering