Andrew Horsburgh
Background, current projects, and where this is heading.
I've spent most of my career in environments where systems fail expensively — aviation, manufacturing, industrial automation, drone operations. Different industries, same underlying problem: complex systems that need to work reliably under real-world conditions, managed by people who can't afford to get it wrong.
That background shaped how I approach building things. I default to operational thinking — what happens when this breaks, who's responsible, how does it degrade gracefully. I think in systems, not features. I plan for the failure mode before I plan for the happy path.
I build production systems using AI as an engineering partner. Not as a code generator or an autocomplete — as a collaborator that handles syntax and pattern application while I handle architecture, specification, quality evaluation, and the domain expertise that determines whether a system actually does what it's supposed to do.
This way of working is sometimes called AI-assisted development, and it's quickly becoming the standard for how software gets built.
The skill isn't using the tool. The skill is knowing what to build, specifying it precisely enough that an AI can execute it, and evaluating whether the output meets the bar.
CipherRank
In App Store reviewA gamified CompTIA exam prep platform — three certification tracks covering Security+, Network+, and the new SecAI+ exam, with 648 missions and 2,242 decision points. It has an AI generation pipeline with two-stage validation and cost-modelled token economics, built so the product works with or without the AI layer.
View case study →BLEKit
In active developmentA two-app Bluetooth security suite — one for professional site audits, one for counter-surveillance. Shared scanning engine, threat scoring, two distinct interfaces for two distinct user profiles.
View case study →Codeword Pipeline
Running in productionA 6-stage real-time ML system that monitors live audio, transcribes speech, classifies content, and extracts structured data. Runs unattended — cron-scheduled to contest hours with automated capture archiving. Over 500 targets identified, six false positives, all caught by the pipeline's own verification stage. Zero application failures.
View case study →Each project pushed a different boundary: AI product economics, monorepo architecture, real-time ML classification. They were all designed and built with AI as an engineering partner.
That entire body of work started with a Google search about speech-to-text about five weeks ago.
The market I'm most interested in is the one that barely exists yet: small and mid-sized businesses who want to run their own RAG pipelines, agentic workflows, and automation systems on infrastructure they control — not rented from a provider who can read every token.
Data sovereignty isn't a feature. It's an architecture decision.
I'm still early in this space. The portfolio above is evidence of how I build, how I think, and how fast I move. The direction is toward helping organisations keep their AI in-house — on their own hardware or on VPS they control — without needing a team of ML engineers to make it work.
If you're working on something in this space — or you're looking for someone who builds this way — I'd like to hear about it.
andrew@adhdigital.caAndrew Horsburgh
ADH Digital
Remote-first, based in Toronto, Canada.
andrew@adhdigital.caOpen to interesting problems — whether that's the right full-time role, contract work, or something I haven't thought of yet.