edwardfalk.com
Electronics engineer · AI systems developer · IT consultant since 2005
I build systems where hardware, software and AI meet: a voice AI living inside a 1991 Macintosh, a wind-plant simulator for the Swedish power grid, and the self-hosted agent tooling I use to run many AI-assisted projects in parallel.
I've run my own IT company, Falk Data, for twenty years: troubleshooting, networks, security and support for companies and private clients. In 2026 I finished a two-year degree as an electronics engineer in embedded systems at Yrgo in Gothenburg, to take that experience somewhere new.
What I bring is the combination: hands-on electronics (oscilloscope, soldering iron, datasheets from the 1980s), embedded Linux and microcontrollers, and a lot of practical experience making AI models do reliable work with tools, memory and review, not just chat.
A Raspberry Pi 5 inside a 1991 Macintosh Classic II drives the original 9" monochrome CRT directly from its GPIO pins. You talk to the AI assistant Happy Mac through a 1950s Swedish Ericofon handset. He runs the apps and games on screen, answers in Swedish or English, and can read live traffic from a passive network tap. He can watch but never interfere.
A deterministic simulation of a small Swedish wind plant. It estimates what the plant would have produced and earned at a real site, and rehearses Svenska kraftnät's mFRR/aFRR prequalification, the test a plant must pass before selling balancing power to the grid. Every number carries an evidence level, from assumed to fitted to real logs.
Private for now
A git-backed memory vault shared by every AI coding session. A nightly "librarian" agent files and links new notes, and an MCP connector on my own VPS makes it reachable from anywhere. Plans get a second-model critique and code gets blind parallel reviews before every push. Earlier work: RAG with Qdrant and Redis, knowledge graphs, and custom MCP servers.
Private: personal data
Says the colour of a garment out loud, built for a visually impaired client. It turns on the flashlight, classifies three frames and reconciles them before speaking, because one frame under household light isn't reliable enough to tell someone who can't check.
In testing
A browser gliding game with real aerodynamics, procedurally generated terrain and thermals to find. The flight model is parameterised so it can fly on other worlds. It's also my testbed for fully AI-generated code.
An Arduino offloads 7-segment display driving to an FPGA over SPI, with and without metastability protection. Also from the degree: a digital guitar filter on FPGA, an autonomous Arduino car and a servo-driven drum machine.
A rhythm-driven space shooter, made as an experiment in how far AI coding can go without my writing a line of code myself. The code, graphics and animation are all AI-generated.
2005 – present
IT support, troubleshooting, networks, security and training for companies and private clients, and more recently AI automation. Sole trader until 2023, then a limited company. The job is problem-solving: understanding what the customer actually needs, then making it work.
2025 – 2026
Built Project ScrLk, from first prototype to the degree-presentation demo.
2024 – 2026
Two-year higher vocational degree. Analog and digital electronics, VHDL/FPGA, embedded C/C++, signal processing, test automation, embedded Linux, agile project work.
Earlier
Informatics, databases, psychology, theoretical philosophy, business.
I play guitar and synth in the Gothenburg band the Blindness and I'm building my own synthesizer.
Fun fact: I was on my way to becoming an artist when I took an internship at a computer shop. Two years later I bought the shop, for almost exactly what I'd won playing online poker.