Mikko Tarkiainen

AI Systems Engineer · AI Architect

From idea to implementation.

AI architecture, put into practice.

I design and build agentic systems with business rules, permissions and testability built into their structure. My architecture work includes writing the code and verifying how it behaves.

Finland · Remote roles, selected freelance projects and software partnerships

Foundations of an agentic system

Control

What an agent may do — and under which conditions.

Evaluation

Does the system behave within its specified boundaries?

Evidence

What was observed, verified and left uncertain.

Conceptual model — not a running system

01 / Selected work

From architecture to implementation.

Four projects. One approach: explicit boundaries, testable decisions and verifiable observations.

Agent Runtime Controls

ACS Guardrail Demo

Problem. An agent’s proposed tool action needs explicit authority before it runs.

Implementation. Deterministic runtime controls, signed and scoped capabilities, and human approval bound to the tool action.

Runtime controlsScoped capabilitiesAudit records

Control Evaluation

Agent Control Evals

Problem. Reported allow/deny decisions alone do not establish what a system actually did.

Implementation. Reproducible evaluations separate policy decisions from observed effects. Explicit specifications, deterministic oracles and adversarial scenarios.

Evaluation harnessesMutation testingDeterministic oracles

Verifiable Execution Receipts

Execution Evidence Demo

Problem. Execution claims need verifiable observations — and uncertainty must be preserved.

Implementation. Signed receipts bind actor and action to hashes of supplied before-state, after-state and result observations.

TypeScriptEd25519Linked receipts

Dealer-Controlled Digital Sales

Kopilotti Sales

Problem. A digital automotive salesperson must move the conversation toward a sale — on the dealer’s terms.

Implementation. The language model converses. Deterministic server-side rules control pricing decisions and sales-flow transitions.

Digital salesDeterministic rulesMagic Link

02 / Expertise

See the system. Build it, too.

Roles where architecture and hands-on implementation belong close together.

AI & agent architecture

Responsibilities, tools and operating boundaries built into the system — not just the prompt.

Permissions & deterministic decisions

Explicit authority and business rules kept separate from language-model reasoning.

Evaluation & reliability

Reproducible tests, observed effects and coverage gaps made visible.

Business problems to working software

From understanding the problem to implementing a solution. Engineering decisions have a commercial context.

Languages & application development

TypeScript · JavaScript · Node.js · Python · Rust · Solidity

Data & infrastructure

PostgreSQL · Redis · Kafka · Docker · WebAssembly

Testing & delivery

Playwright · Jest · Vitest · GitHub Actions

03 / About

Own the problem. Answer for the outcome.

Mikko Tarkiainen

I am an AI systems designer and builder based in Finland. Two decades as an automotive entrepreneur shaped how I work: own the problem, make decisions under uncertainty and remain accountable for the outcome.

Today, I design and build AI systems with explicit operating boundaries. I combine an entrepreneur’s practical judgment with an engineer’s need to understand what a system actually does.

Engineering principle
Flag, don't hide.

Distinguish what was decided, what was observed and what was verified. State what remains uncertain. Good architecture does not hide these differences.

04 / Contact

From architecture to working code.

I am looking for roles and collaborations where design comes with responsibility for implementation. I am not seeking roles limited to advisory work, presentations or architecture documentation. If you need someone who designs, builds and tests, let’s talk.