Tomi Linkinen - Generative AI Systems Designer Portfolio

Generative AI Systems Designer bridging architecture, requirements, and stakeholder communication for cutting-edge R&D.

About Me

During my university studies, I took on as many team projects as I could. I quickly noticed a recurring pattern: teams consistently needed someone who could handle architecture, requirements, research, and stakeholder communication all at once. I made a conscious choice to fill that gap. I focused my efforts to understanding how the code gets built and how it can be deployed so I had enough understanding to talk with programmers fluently, but I dropped memorizing syntax and heavy programming tasks. Then, the rest of the time I went all in on everything else a software project needs to succeed because I truly believe that brilliant code cannot save a poorly designed system, and that failed communication can ruin a brilliantly coded and designed project.

After graduating from Tampere, I took a position in Oulu on the Mastering Generative AI initiative. That is where I saw how naturally design and team leading connect in practice. When our supervisors introduced a broad vision, I was able to turn that vision into implementation plan, solving the "how" by designing the system, diagramming the architecture, and writing the requirements. I could then carry that design to stakeholders and to the team to collaborate on and continue working on the project serving the teams needs as, be those needs as sprint planner, researcher, presenter, or anything else that was needed.

Experience

Mastering Generative AI Initiative - University of Oulu

Full-time employment, 2025–Present

Responsible for system design and software architecture. Translating frontier R&D initiatives into roadmaps, defining requirements, and adjusting systems to fit real-world constraints. Navigating ambiguous frontier R&D through clear communication. Facilitating team operations using Lean and Agile methods to organize workflows and drive projects from zero to one. Devising use cases, workarounds, and solutions to cutting-edge generative AI challenges.

  • Portable Local RAG POC: Architecture, Design & Roadmapping
  • Edge IoT Project: Architecture & Design
  • AI Workstation: Hardware Proposal
  • Project Management: Acted as team lead and project manager. Onboarded team members, brought in Agile/Lean methods and development process, and drove 0-to-1 software delivery.
  • Hybrid Agentic IDE: Combining Antigravity, Kilocode & DGX Spark hardware to run local and cloud llms together to work in single codebase without swapping developing platform.

Master of Science, Computing Sciences - Tampere University

Graduated 2024

Specialized in Human-Technology Interaction. Studied design of software and user interface, how humans experience software, how generative AI can be utilized by the end user and in software development, and how the end-to-end software process works in practice. Delivered real client software projects and innovation efforts. Managed multi-phase development cycles using Agile and Lean.

  • Master's Thesis: Generative AI in game development. Built a working UE5 prototype using AIVA, Stable Diffusion, Midjourney, and LLMs as the research artifact. Grade: 4/5.
  • Solita: UX research and Figma prototyping for Nysse and HSL public transport focus on accessibility enhancements.
  • Gofore: Innovation and research project: Drone and 3d environment assisted crime scene investigation methods.
  • Valmet: UX research for a customer-facing driver download service.
  • Huld: Mobile lunch time planning application for office workers with Google Maps integration. Project manager and designer.
  • TAYS Children's Hospital: VR activation game proto.

Bachelor of Social Sciences, Multidisciplinary Communication - Tampere University

Graduated 2021

Studied interactive media and communication. Structured the degree as a deliberate split between computing sciences and multidisciplinary communication.

  • Bachelor's Thesis: "Benefits and Drawbacks of Esports in an Educational Setting". Grade: 5/5.

Competencies

  • Systems Design: Prototype (Figma), Validate (Miro), Document (Confluence), Research (Lean UX), Architect, Diagram (Draw.io).
  • Project Execution: Iterate (Jira), Analyze, De-risk, Unblock, Orchestrate (Agile/Lean/Kanban), Roadmap.
  • Generative AI: Remix (Multimodal Generation), Track research, Geopolitics of AI, Scrutinize outputs, Navigate models (Gemini, Claude), Prompt engineering.
  • Local AI: Inference (llama.cpp, vLLM), Quantize (GGUF), Select (SLMs), Multimodal (Whisper, Stable Diffusion), Build (RAG), Pipeline (FAISS).
  • Agentic AI: Engineer (Antigravity), Combine (Kilocode), Manage (DGX Spark), Source (HuggingFace), Design architecture, Prompt engineer.
  • Stakeholder Communication: Record, Align, Produce (Premiere, Cubase), Translate, Present, Create (Photoshop).

Project Deep Dives

Portable Local RAG POC

Goal: Develop a proof-of-concept, local, portable, offline, and secure Retrieval-Augmented Generation (RAG) system that runs on a standard laptop to act as a writer based on user documents.

Challenges: Running LLM inference locally on standard hardware, portability constraints, data deprived environments, and variable document structures.

My Role: Architecture, design, requirements, and solving constraints. Project management using Lean/Kanban via Trello.

Solutions: Balanced hardware limitations by allowing more generation time (fire-and-forget system) and using a 14B Phi model. Utilized quantized GGUF models with llama.cpp to run on CPU/RAM. Portability solved via USB-C. Chunk filtering for data deprivation. RAG output handled via prompt engineering. System designed so USB-C retained no session information for security.

Edge IoT Automation

Goal: Develop a proof-of-concept, scalable, and edge-compatible AI solution using GenAI technologies to address IoT challenges.

Challenges: Unpredictable multi-hardware environment, multiple protocols, runtime constraints on local execution, scaling, and data retention during fail cases.

My Role: Sole architect and designer. Innovated novel solutions, designed technical feasibility, architecture, LLM orchestration, and researched new technologies.

Solutions: Discovery layer (Adaptive pattern matching for IoT identification), Code Scaffolding (Template-based code generation), Session State Preservation (Persistent management with fail-safe recovery).

Local AI Workstation

Goal: Plan a complete hardware component list for a local AI workstation capable of heavy matrix multiplication with remote capabilities and low noise levels.

Challenges: Compatibility between workstation and server infrastructure (form factors, EEB constraints, cooling), thermal vs acoustic limits, preventing GPU sag, and beating OEM cost.

My Role: Solo project designing and selecting hardware.

Solutions: Used Gigabyte MS03-CE0 (LGA4677 ATX motherboard) to fit standard chassis. Prioritized CPU thread count and L3 cache for GGUF model CPU inference. Designed a bottom-to-top positive-pressure airflow system in Antec Flux Pro with Noctua fans. Storage optimized with 8TB U.2 SSD for scratch space and isolated OS drive. Over-provisioned power delivery with 2200W PSU.

Master's Thesis: Generative AI in Game Development

Goal: Explore how generative AI could be utilized by developers as a practical tool for multi-modal content (code, writing, 2D graphics, audio) in game development using an Unreal Engine 5 prototype.

Challenges: Novelty of field with lack of academic literature, hardware constraints for local inference vs cloud APIs, limiting scope, and managing UE5 complexities.

My Role: Full lifecycle owner, generating 15,000+ images for evaluation, building the game prototype, and researching workflows.

Solutions & Findings: Developed hybrid pipelines (e.g., AIVA MIDI to Cubase to avoid audio artifacts, Midjourney to Photoshop/Stable Diffusion for precise image control). Identified LLM limitations such as lack of subtlety, context window constraints, and "style drives content" dilemmas in image generation.