For more than 20 years I have been working with AI. My passion is helping organizations turn modern AI — and especially agentic AI — into shipped products, working processes and decisions that matter.
The arc: software-agent research at VTT (2001+) and Aalto/TKK, retail forecasting at unicorn-scale RELEX (CTO), the HUS Data Lake cloud-AI platform at TietoEVRY, AI in international tax at Aibidia (early LLM & RAG), and the NLP→LLM platform renewal at TalentAdore. Today: Head of AI Development at Siili Solutions and PhD research on software agents and LLMs at Aalto University.
Day to day I am a heavy user of frontier LLMs (OpenAI, Anthropic, Google) and the leading agentic coding tools — Claude Code, OpenAI Codex, GitHub Copilot Pro and Cursor. Two working principles guide the practice:
In my PhD at Aalto I dig into the internals of modern LLM architectures — to understand exactly where Transformer-based AI starts to bend, so that benchmarks, agent designs and product decisions match the real model limits rather than the marketing.
Behind the AI work sits 15+ years of senior technical leadership — 4× CTO across four sectors (retail unicorn RELEX, healthcare Hifian as founder, tax Aibidia, HR TalentAdore), scaling teams, services and SLAs through the toughest international growth phases.
EXPERTISE:
Heading AI Development for Siili's Digital Engineering Business Line. Setting the AI strategy and delivery motion for client engagements across retail, healthcare, taxation, energy and HR.
Continuing the deep hands-on use of coding agents and frontier LLMs described under AI Principal below — but now with broader BU-level scope: turning that agent-driven engineering practice into a delivery capability that Siili clients can buy and deploy.
AI Principal at Siili Solutions, bridging business expertise with AI and GenAI. Applying data science methods and modern, well-structured ways of working — hands-on at the keyboard, not slideware.
Extreme, daily use of coding agents and frontier LLMs — Claude Code, OpenAI Codex, GitHub Copilot Pro and Cursor, paired with frontier models from OpenAI, Anthropic and Google — building prototypes, internal tools and PoCs across multiple client domains. The work doubles as a lived case study for the agent-driven engineering practice I now help others adopt.
Currently the largest AI token consumer across the Siili Group (~1 000 employees) — using that bandwidth to stress-test agentic delivery in real client engagements and feed the learnings back into the organization.
PhD research at Aalto University on software agents, large language models and generative AI — connecting the early multi-agent line I started at VTT to today's LLM-driven agentic AI.
Hands-on daily with frontier LLMs, defining Agentic LLM benchmarks, and building practical multi-agent systems. The research line focuses less on toy demos and more on how real agents are measured, how they coordinate, and how they actually deliver value in production-grade contexts.
Part of the work goes into the internals of modern LLM architectures — studying exactly where Transformer-based AI bends and breaks, so that benchmarks and agent designs are calibrated against real model limits rather than marketing claims.
Status (May 2026): Pre-examination comments received; finalizing the manuscript by June 2026; public defense and approval scheduled for Q3/2026. Five conference papers published in H2/2025, including a Best Paper Award at PCSI 2025 on AI agents and a 2nd Best Paper Award at CHIRA 2025 on LLMs and agents. Over 25 international publications across the wider AI / advanced-analytics line.
CTO at TalentAdore — a recruitment-tech SaaS that adds a human touch to hiring through advanced AI.
Led the renewal of the company's core SaaS technology — migrating from an in-house, proprietary NLP stack to a public-cloud, state-of-the-art LLM stack. That move unlocked GenAI-grade capabilities for recruiters while collapsing operating costs and time-to-feature for the product team.
Implemented LLM features on AWS for hundreds of production customers processing personal-information data, with the team deploying advanced GenAI design patterns such as Retrieval-Augmented Generation (RAG) into production use.
Designed and implemented a multi-tenant data analytics platform on AWS with a column-oriented data layout for high-performance OLAP — 15× improvement in query speed compared to the previous analytics solution.
Centralized R&D into the Helsinki region (~15 on-site employees, transformed from a distributed outsourced R&D org) and hired the key people behind the new platform — building a focused, senior-loaded team that could ship on the new stack at the pace the market demanded. Reported technology strategy and R&D outcomes to the CEO and Board.
CTO at Aibidia — bringing AI to international tax for global enterprises. Led the cloud platform that helps large multinationals digitalize their transfer-pricing processes through machine-learning and AI algorithm implementations.
Designed the platform's DAG-formatted function graphs — explicit, auditable computation graphs that capture transfer-pricing logic as composable transformations across millions of intercompany transactions and entity hierarchies.
Drove an early production deployment of large language models (LLMs) and Retrieval-Augmented Generation (RAG) — well before either became mainstream — to turn tax legislation, customer documents and audit trails into AI-assisted, searchable workflows for tax professionals.
Business impact: SaaS revenue grew 30×, workforce expanded 6× to ~60 employees, R&D organization of ~35 people transformed the product from PoC to a robust SaaS platform sold to some of the largest enterprises in the world. Reported technology strategy and R&D outcomes to the Board; contributed to the Series A funding round (€13 M).
Security & compliance: led the projects to obtain ISO 27001 and SOC 2 / ISAE 3000 Type II certifications. Built the platform's analytics on Microsoft Azure with a full BI stack (SQL Server, Data Factory, Power BI, Pyramid Analytics).
Headed Cloud Analytics and AI at TietoEVRY Healthcare. Responsible for the HUS Data Lake — a Microsoft Azure–based public-cloud AI healthcare platform integrating all main HUS systems (Apotti / Epic plus 50+ supplementary systems) into a single research- and care-grade data layer.
The HUS Data Lake is the largest healthcare data platform in Finland, covering roughly 40% of the country's population and hosting on the order of 800 million laboratory results and ~1 million imaging records. It is the substrate for the 360° Patient view and a steadily growing fleet of AI and analytics use cases serving HUS clinicians and researchers.
Created accurate Covid-19 forecasts when the first corona wave hit Finland — featured by TietoEVRY: corona-epidemic-prediction-accuracy is top-notch.
Participated in RFPs up to 100 M€, defining solution architecture on top of public-cloud technologies for international healthcare customers.
Senior AI role in the Tieto / TietoEVRY CTO Office, reporting directly to the CTO through the period when Tieto and EVRY merged into a ~24 000-employee company. A landscape position covering large sales cases, project risk management, and internal and external transformations through the merger.
Drove data- and AI-driven transformations for large Nordic healthcare and retail customer engagements, working alongside the customer-facing teams and the CTO Office on cloud architecture and AI strategy. Participated in RFPs up to 300 M€+ in total budget.
Senior technical leader through RELEX's unicorn run-up. Personally hands-on in the largest customer deployments, where the SaaS platform calculated dozens of billions of forecasts across global retail networks — supply-chain mathematics at planetary scale.
Organized world-class R&D efforts in partnership with large hardware manufacturers, pushing the limits of in-memory and GPU-accelerated computation for forecasting and optimization. Led the mathematical AI algorithm development behind the forecast / replenishment / workforce-planning accuracy that customers depended on.
During the 2015–2019 run, RELEX grew from ~100 to ~600 employees and my organization from ~30 to ~150 across 17 technical teams in 11 international locations — development, QA and technical delivery of 200+ international projects. Constructed an enterprise-level cloud service with on the order of ~300 000 GB RAM and ~20 000 processor cores. Built the SaaS & IaaS platforms that run enterprise services for the largest international retailers (Coop Denmark, Morrisons, Rossmann and others), with the SLAs, service models and technical-leadership patterns behind them.
Saw the full customer lifecycle — from early sales through implementation to continuous support — hundreds of times in international Champions-League competition, and the full technology-stack evolution from application code down to physical infrastructure.
Scaled the technology and the organization as RELEX grew from ~30 to ~100 employees. Owned the end-to-end technology line — development, QA, agile delivery and 24/7 customer support — across every market the company operated in.
Deepened the in-memory database (IMDB) forecasting and replenishment algorithms at the heart of RELEX's retail planning suite. Worked hands-on across mathematical algorithms, data science and machine-learning models that powered the forecast / replenishment / workforce-planning output customers depended on.
Beyond the management line, I regularly embedded directly with the largest international customers — closer in spirit to a forward-deployed senior engineer than a traditional director — taking responsibility for the toughest performance, architecture and analytics problems that came in from the field.
Built the SaaS & IaaS platforms that run enterprise services for international customers (Coop Denmark, Morrisons, Rossmann and others) and set up the SLAs, new teams and new technical approaches that supported the growth.
Senior leader at RELEX as it grew from a one-team startup to ~30 people. Led the technical delivery of the largest customer engagements while the platform was still maturing.
Drove product expansion — building capabilities pulled from real customer needs into the core SaaS, including the in-memory database (IMDB) calculation engine and retail forecasting algorithms that became the technical core of the company's later international growth — and sparred closely with the small product-development team on architecture, performance and database design.
Developed Function Graphs over BOM (Bill-of-Materials) structures as a composable engine for forecasting — letting the team generate new types and classes of forecasting algorithms by re-wiring calculation graphs on top of the product/component hierarchy rather than hand-coding each algorithm. The pattern became a long-lived primitive in the platform.
Recruited the first DevOps team and stood up the SaaS operations, performance tooling and integration patterns that the platform would lean on through the next decade.
Co-founded a part-time startup focused on marine LNG (liquefied natural gas) ship conversion. The team of three won the 1st prize in an innovation competition (€100 k) and secured funding for a €10 M LNG ship-conversion project for a customer.
The 2014 oil-price decline abruptly shifted shipowner investment appetite for LNG conversions; the venture was ramped down accordingly.
CTO, Head of AI and co-founder at Hifian. The work combined software agents, analytic business-intelligence applications, decision-support systems and data science, applied to national-scale clinical and financial healthcare data.
Main architect of the Finnish National Special Care cost database. Designed the platform that ingests cost-per-patient records from every Finnish hospital and produces combined cost weights across the diagnostic, organizational and time dimensions at multiple aggregation levels. The output drives benchmarking, invoicing and management decision support for special-care units nationwide.
Led the project that standardized how Finnish hospitals calculate costs-per-patient and invoice the parties covering them (final deliverable: ~100-page report). On top of this, implemented the analytical calculation engine that combined all special-care unit costs with clinical data to produce the price lists ("cost weights") used in invoicing.
Defined new types of flexible N-dimensional affinity functions used to analyze hospital cost-per-patient data — a composable similarity primitive that let analysts pivot across patient, diagnosis, treatment, organizational-unit and time dimensions simultaneously, surfacing comparable patient cohorts and cost structures across hospitals that traditional fixed groupings missed. The results were shared at international conferences (PCSI, ICEIS, ESTSP, EDBDT) and in the Finnish Medical Journal (Lääkärilehti).
Contributed to Nordic-level DRG development (Finland, Sweden, Norway, Denmark, Iceland, Estonia), analyzing the financial and medical impact of proposed changes — https://www.nordcase.org/
Postgraduate research at Helsinki University of Technology (now Aalto). Completed my Licentiate of Technology in 2006 at the intersection of software agents, multi-agent systems and applied AI, building on the agent-research line started at VTT.
Across 2005–2013 the work extended into natural language processing, information retrieval and data mining, with 10+ peer-reviewed international publications and conference presentations across Europe. Applied-research engagements brought these methods into healthcare (clinical & financial analytics, DRG / cost-per-patient systems) and retail (forecasting, replenishment, planning) contexts — the same domains I later operationalized at scale at Hifian, RELEX and TietoEVRY.
Healthcare IT Systems, Product Development.
Participated as an architect in implementing cost-per-patient systems that define bills that hospitals send to parties that pay special services in Finland (billions of euros per year). The system was based on two dimensional statistical grouping to define homogenous groups in clinical and financial dimension ("DRG, Diagnose Related Groups").
Field: Data Warehouse projects, Extract Transform Load (ETL) processes, Diagnosis Related Grouping (DRG) systems.
Software Agent Researcher at VTT, studying applied AI and machine learning with a focus on software agents operating in mobile, context-aware environments — agents that sense the user's situation (location, device, connectivity, task) and act on their behalf. Both my Master's Thesis and Licentiate's Thesis (each graded 5/5) explored software agents and multi-agent systems as the design pattern for this kind of adaptive software — the intellectual ancestry of today's LLM-driven agentic AI.
Our research group published actively in international academic forums — peer-reviewed conferences and journals on software agents, multi-agent systems and applied AI, with travel and presentations across Europe and beyond.
From that vantage point I concluded that the practical future of adaptive AI agents would not come from explicit symbolic reasoning but from flexible algorithms that learn directly from data. That conviction steered my later work toward data science and natural language processing (NLP) — and, two decades on, the LLM-driven agent stack has validated both threads: agents as a design pattern, learned models as the substrate.
My first industry role. Developed PKI (Public Key Infrastructure) software for payment cards as part of Setec's smart-card programme — the cryptographic plumbing that lets a chip card authenticate itself to terminals, banks and clearing networks.
The work happened inside Setec's high-security-classified Banknote Printing and Bank of Finland departments. Setec had been spun out of the Bank of Finland Security Printing House in 1993, and the combined facility still served as Finland's banknote printer for the central bank — a tightly controlled environment with the access protocols, segregated networks and clearance procedures that come with printing legal tender and personalising bank-grade chip cards.
Setec was acquired by Gemplus in 2005, merged into Gemalto in 2006 (Axalto + Gemplus), and ultimately became part of Thales through the Thales–Gemalto deal that closed in April 2019. The Finnish operation continues as Thales DIS Finland Oy.