Building AI systems and software tools end-to-end, from multi-agent pipelines and LLM integrations to Python backends, REST APIs, and automation workflows. Portfolio projects with public deploys, CI-backed pipelines, and paying client delivery.
I build AI systems and software tools end-to-end, from multi-agent pipelines and LLM integrations to Python backends and automation workflows. My background in cybersecurity shapes how I approach systems: I think carefully about failure modes, reliability, and what happens when things run unsupervised.
Currently completing my MSc in Information Technology at the University of Turku, specialising in Robotics and Autonomous Systems. Master's thesis planned; topic to be agreed with my supervisor.
Languages: Finnish (native), English (professional), Norwegian (elementary, improving).
4-agent LangGraph pipeline for the AMD Developer Hackathon 2026: MITRE ATT&CK techniques to Sigma rules, SOC response guidance, and validation scores. Deployed on Hugging Face Spaces (demo mode by default for reliability). Live inference path with vLLM on AMD ROCm exists; Qwen verifier sidecar audits pipeline output.
Portfolio RAG code and security auditor: Python snippets to OWASP-mapped scores and citations grounded in 20,875 embeddings from 8 security books. FastAPI, pgvector, Redis, LangGraph ReAct agent. pytest suite, GitHub Actions CI/CD, and a 15-case eval runner (7/15 passing, failures documented). LoRA fine-tune artifacts in a companion model repo.
Multi-agent system for Terraform and operational metrics analysis. Deterministic config and metrics agents feed a Claude supervision layer that synthesises drift, waste, and optimisation opportunities. Dockerized UI replaced the original Streamlit prototype.
Fully local voice assistant built from scratch using faster-Whisper for speech recognition, Ollama for LLM inference, and Piper TTS for voice output. No cloud APIs. Includes a FastAPI web interface with browser microphone support and custom fine-tuned model support via Ollama Modelfiles.
Paying client delivery: internal production system linking work orders to pipe and material inventory. Completing a job deducts stock, tracks remnants, and logs a full audit trail. FastAPI backend, React UI, PostgreSQL, nginx and systemd on Linux. Auth, CSRF protection, health checks, and handover documentation included.
Attack-surface database and import/query pipeline with scan snapshots and diffing. Integrated HexStrike into automated security analysis via MCP in Cursor. Built the SQLite pipeline and lab workflows; not the upstream HexStrike AI framework.
End-to-end Rock Paper Scissors League in TypeScript and Python: API integration, persistent storage, Server-Sent Events for live state, and clear backend architecture across two stacks.
Local-first tools with a shared design system: Ledger and Debts (accounting), Career job-search dashboard, PdfConvert (PDF to Markdown for AI ingest), and VideoDownloader (local yt-dlp). React and Python apps with systemd/local HTTPS where needed.
Browser extensions, API utilities, and earlier lab work. Full index on the projects page.
Designing and building AI-powered systems and backend workflows with a focus on practical applications and local-first architectures. Developing modular pipelines integrating LLMs, APIs, and structured data processing. Building end-to-end projects including voice interfaces and automation platforms.
Regulated healthcare environment in client home care. Responsibility, discretion, and clear communication in Finnish with non-technical clients. Not healthcare IT; strengthens reliability and customer-facing work.
Building and maintaining web-based applications and developer tools, including browser extensions and full-stack projects. Focus on API integration, system design, and clean, maintainable implementations.
Python tools for data processing, monitoring, and workflow automation. Structured data pipelines and system-level integrations.
Log analysis, documentation, and security monitoring support (2-month engagement).
Analyzed system architectures and networked environments to identify weaknesses and improve reliability. Built scripts and tooling to support analysis workflows and system validation.
Conducted independent analysis of web systems, focusing on identifying implementation issues and validating real-world impact through proof-of-concepts.
Earlier experience and additional roles available on LinkedIn.