5K–10Krequests per second · ~600M / day
Spring Boot services I maintain at Klarna against very large databases — adapting to data-provider changes without downtime. The altitude where mistakes become incidents.
Klarna · OSIS Customer API · 2024–2025№ 01
Other receipts on fileSix numbers I'm willing to defend.
2M+conversations / month
AWS pipeline through OpenAI Whisper and ChatGPT, three regions, throttled, fault-tolerant, monitored.
Klarna · 1st-line tooling · 20245countries
Recovered an "unsupportable" 3-year-old Android codebase and shipped it across five country variants.
Telia Sonera · 2015–2016~50apps shipped
15+ years of native iOS and Android. Custom UI, NDK libraries, React Native bridges, embedded patches.
Career portfolio · since 20084years rebuilding Office in .NET
Co-architected Syncfusion equivalents of Excel, Word, Visio, charts, reports — 100% managed code.
Syncfusion · 2005–200920+years commercial IT
From WDM drivers and x86 assembler in 2001 to Knowledge Graphs in 2025. Both ends of the stack.
Since 2000№ 02
What I'm hired forThe six shapes of work I'm actually brought in for.
- 01Production systems at scale.Klarna's Spring Boot at 5–10K requests per second. AI pipeline at 2M+ conversations per month across three AWS regions. Load that's senior territory.
- 02Migrations that have stalled.JS → TS. Java → Kotlin. GWT → native. Webpack 4 → 5. Monolith → monorepo. Modernization that ships, not theater.
- 03Codebases the original team has left.Recovered Telia's three-year-old Android repo and shipped it across five country variants. Drove TrueCaller's monthly crash rate down by a factor of one thousand.
- 04Greenfield that needs senior judgement.From a blank repo: architecture, vendor selection with evidence, CI/CD, the first three modules — then a clean hand-off. EasyPark, KOHLs, AutozvukUA, and a dozen mobile apps did exactly this.
- 05AI integration that's actually grounded.Knowledge graphs, RAG, fine-tuning on real expertise. Production AI grounded in the actual semantic structure of your business — not bolted-on chatbots and demo magic. The thinking is in the bylines below.
- 06DevOps platforms with AI-enabled velocity.Build pipelines, release management, observability, deploy automation. Currently turning If Insurance's old-school Guidewire deployments into AI-enabled, high-velocity pipelines. All DevOps key metrics in scope.
I guide product over the reefs of migrations and scaling.
— positioning, in one line
№ 03
Bylines · ITNext · MediumActive dispatches from the build-out of Big AI.
I write on AI strategy and the operational realities of building with it — for the engineers and architects doing the integration, and the leaders who hire them. The first four pieces (April–May 2025) form a connected manifesto on Big AI; two later dispatches extend the argument into token-economics and design.
- 01Featured · ITNext · Apr 2025 · 12 min readRead on ITNext →
Beyond Big Data: The Rise of Big AI.
The fifty-year era of data-driven decision-making is ending. Big Data was built to describe what already happened — observation, not action — and most businesses now have more information than they can metabolize. Big AI inverts the relationship: from mirrors to headlights, from hindsight to foresight, from collection to autonomous response. The piece frames the shift architecturally: what changes when intelligence, not data, becomes the substrate underneath the systems we build.
- 02ITNext · May 2025 · 12 min read · Strategy → executionRead on ITNext →
The Hidden Layer of AI That Most Teams Overlook.
Most organisations don't have a system of intelligence — they have a system of intuition. Tribal knowledge, undocumented decisions, Slack threads, the one engineer who "just knows how it works." The piece argues that the missing operational layer is rationale itself: the why behind every requirement, ticket, and architectural choice. AI doesn't thrive on tasks; it thrives on reasoning. In this framing, the why becomes the new source code — and the layer most teams forget to build.
- 03ITNext · May 2025 · 10 min read · Make knowledge work · Adjacent to Klarna receipt ↑Read on ITNext →
Building a Knowledge Graph for Business: The Semantic Backbone of Big AI.
A practical step-by-step on building the semantic layer that grounds enterprise AI: ontologies, linked data, SPO triples (subject-predicate-object), and LLM integration. Where most teams reinvent the wheel from scratch with brittle prototypes, the article argues the foundations have already been built by the open-source and research community — what's needed is the structure that converts ambient business knowledge (Slack threads, Jira tickets, meeting notes) into a dynamic, queryable layer AI can actually reason against. Adjacent to the Klarna Knowledge Graph receipt above.
- 04Medium · May 2025 · 11 min read · The case for hiring moreRead on Medium →
Think AI Means Fewer People? Here's Why the Smartest Companies Are Hiring More.
A direct argument against the most common AI strategy of 2025: replacing experts with chatbots. AI is an amplifier — it scales what you feed it. Fire the people whose expertise makes your company worth running, and you scale their absence. Off-the-shelf LLMs hit 70–80% accuracy on simple tasks but fall over on the decisions that actually matter — code generation, strategy, regulatory review. The fix isn't more tokens. It's fine-tuning on your best people's reasoning. Which means hiring more experts, not fewer.
- 05ITNext · Nov 2025 · 19 min read · Investment + scaling thesisRead on ITNext →
The 1,000 Token Wall: Why AI's Speed Limit Is About to Shatter.
Why are NVIDIA's latest systems advertised at 72,000 tokens/second while you watch ChatGPT trickle out at 1,000? The article maps the autoregressive bottleneck — memory bandwidth, the KV cache, sequential generation — onto a precise historical parallel: the Pentium-era CPU wall of the 1990s, broken not by one breakthrough but by a stacked set (pipelining, branch prediction, speculative execution). The same stack is now coalescing for AI: FP4 precision, speculative decoding, KV cache compression, processing-in-memory, model orchestration. Conservatively multiplied: 100× throughput. Optimistically: thousands. The product economics restructure on the way down.
- 06Medium · Apr 2026 · 22 min read · Design × AI · NewestRead on Medium →
Why Good Design Feels Right — and How AI Can Learn to See It.
"Clean." "Premium." "Modern." These words say almost nothing — they are aftertastes, descriptions of what people feel after the design has already acted on their perception. The article breaks "feeling right" into measurable primitives — rhythm, contrast, repetition, deviation, focus, meaning, narrative — then turns them into a framework AI can apply: read a UI, evaluate it against the primitives, improve it. The argument cuts both ways. It makes design more rigorous for designers, and it makes design intelligible to systems.
More writing at @olku on Medium ↗ · ITNext ↗
№ 04
Track recordTwenty-five years. Receipts in chronological order.
- 2025 → presentIF INSURANCEDevOps platform migration — old-school Guidewire deployments to AI-enabled, high-velocity pipelines · release management, planning, build-time optimization · all DevOps key metricsSr. DevOps
- 2022 → 2025KLARNA⟳ ReturnSpring Boot services at 5,000–10,000 requests/second · GraphQL redesign · AI pipeline (2M+ conversations/month, 3 AWS regions) · Knowledge Graph from codebase via RAG/MCPSr. Full-Stack
- 2022 → 2023DESIFERTech Lead for Deutsche Telekom — small team, three POCs shipped, next-gen home entertainment infraTech Lead
- 2020 → 2022SPOTIFY⟳ ReturnClient Platform · Antipasta · Migrations WS · Android Devs Community · Java→Kotlin, modularizationAndroid · SM
- 2018 → 2020KLARNAReact Native monorepo · iOS/Android release management · native bridge modules · stability + 3rd-party SDKsSr. Mobile
- 2016 → 2018EASYPARKAndroid app from zero · MVP · GPS · payments (SEPA, PAYPAL, STEX) · GWT→native migrationSr. Android
- Mar → Jul 2016SPOTIFY5-month engagement on Premium UX squad — first 100 days improvement push · bug fixes, analytics, statistics-led problem solvingSr. Android
- 2015 → 2016TELIA SONERARecovered 3-year-old "unsupportable" Android codebase · 5 country variants · static-analysis-driven refactorSr. Android
- 2014 → 2014TRUECALLERDrove monthly Android crashes from 1,500,000 to 1,500 · CI/CD from zero · TDD/BDD adoption · Scrum coachingTech Lead
- 2005 → 2014SYNCFUSION & 50+ APPS4 years rebuilding Office in .NET (Excel, Word, Visio, charts, reports, source-code editor) · 50+ mobile apps · embedded Android · e-commerceArchitect
- 2000 → 2005ELEKS · INTEROBJECTWDM drivers · ASN.1 codecs in C · x86 assembler · ActiveX · Jersey Island vehicle registration systemEngineer
№ 05
Get in touchI make itO.K.
A thirty-minute call, no slide deck. We discuss the work, agree whether I'm the right person for it, and if so — we scope it together.