theopscrafter.com
Responsible AI Product EngineerBased in Finland · EU privacy & security by design

I build AI-assisted finance tools
— with the AI kept on a leash.

View projectsGet in touch
Responsibility ledgerWho owns what
CODE
Deterministic code owns the moneyTotals · VAT · dates · dedupe · postings
LLM
LLMs earn the languageCategories · explanations · Q&A drafts · research
HUMAN
Humans approve everything that mattersReview · filing · merges · final posting
01

How I work with AI

Not anti-AI, not AI-everywhere. Each project below states where AI is used, where it is deliberately excluded, and what guardrails make that boundary hold.

P/01

AI where it earns it

Ambiguity, language, research, drafts — that's LLM territory. Totals, VAT, identity, postings — that's code, rules, and humans.

P/02

Parser of record vs. enrichment

OCR and validation rules own the numbers. The LLM suggests, explains, and cites — and every suggestion is logged.

P/03

Human approves money

Nothing financial posts itself. Low confidence goes to a review queue, and the audit trail shows suggested vs. approved.

P/04

EU privacy and security by design

Built in Finland with EU GDPR principles guiding the architecture: data minimisation, purpose limitation, tenant isolation, encryption, audit trails and fail-closed access. Sensitive decisions remain reviewable by humans.

02

Projects

Product and systems work across finance operations, privacy-sensitive records and deterministic automation.

01 Live — demo on requestCase study

Automated Bookkeeping — multi-tenant finance operations, human-approved

A multi-tenant financial operations platform that takes a business from receipt capture to accountant handoff, with tenant isolation, fail-closed file handling and append-only audit records throughout.

Multi-tenant isolationFail-closed file identity (version + hash)Rules-first, LLM-second Read case study → 02 LiveCase study

FamilyTree App — graph-native heritage platform, AI kept out of the record

A family-tree and heritage platform that helps people of Nordic descent trace, document and share their family history, with dedicated workspaces for heritage associations and their archives.

Graph modeling (nodes, edges, traversals)Policy-guarded multi-tenancyPrivacy by design for sensitive PII Read case study → 03 Private — diagram onlyCase study

Purchase Invoice OCR — inbox to AP system, no LLM in the money path

A private, internal AP automation that moves purchase invoices from an Outlook inbox to the AP system without manual keying. Built on .NET, Microsoft Graph and Azure Document Intelligence, engineered for auditability over cleverness.

Idempotency and exactly-once-ish deliverySchema-first integration (XML contracts)Dead-letter and reconciliation patterns Read case study →
03

Skills

AI — used deliberately

  • LLM APIs as enrichment, not source of truth
  • Grounded Q&A with citations
  • Confidence thresholds + human-in-the-loop
  • AI-assisted development, human-reviewed

Deterministic engineering

  • Azure Document Intelligence (OCR)
  • .NET services
  • Microsoft Graph API / Outlook automation
  • SFTP, XML schema validation
  • Idempotency, queues, reconciliation

Data & security

  • PostgreSQL + Prisma
  • Neo4j (graph traversals)
  • Private object storage / Cloudflare R2
  • Encryption at rest, audit trails
  • Tenant isolation, RBAC

Product stack

  • Next.js / React / TypeScript
  • React Native / Expo, Electron
  • Inngest, Stripe, Resend
  • Vercel, Sentry
  • Signicat (Nordic eID)
04

Contact

[email protected]

Best way to reach me: email. Architecture walkthroughs are available on request.