A Diary of an Engineer

I have a personal rule for technical strategy: never build on top of a legacy system for fun. It is usually a recipe for friction without the reward of elegant design. Yet, this week, I found myself breaking my own rule while finalizing the workflow for a " Tax Bridge " UK project. The goal was simple: solve my own problem. As a frugal solopreneur, I wanted a minimalist user journey. No account registration, no login, no harassment. Just a simple CSV or spreadsheet drop, some basic checks on revenue and profit/loss, and a direct handshake with HMRC and Companies House . I call it data minimization . In practice, it felt more like navigating a three-body problem. The Friction of Legacy Integrating Stripe was a breeze. With code agents, we handled the integration in seconds. But the moment we touched the " Gov-Tech " stack, the pace shifted from light speed to a crawl. HMRC and Companies House operate on their own timelines. They rely on XML Gateways and GovTalk s...

I spent part of this weekend revisiting the predator equation . I came across a video explaining the mathematical game of survival, specifically the Lotka-Volterra equations that model how two species interact in a cycle.  It is a beautiful, oscillating balance: as the prey population grows, the predator population follows with a lag, eventually over-consuming the resource and triggering a mutual decline until the cycle resets. While the video focused on biology, I found myself thinking about global macroeconomics . In my view, the macro environment follows a similar cyclic nature driven by two forces with a strong negative correlation. I began mapping them in my head. Capitalism , led by growth and inflation , acts as the stronger force. The labor market and employment rates act as the base. The central bank sits in the middle, attempting to act as a smoothing influence or a regulator of the oscillation. When two instruments have a correlation near –1, the peak of the stron...

In the early stages of building WorkRightsNow.org , I’ve been constantly reminded of the gap between " code-complete " and " system-ready ." I finally have the module " Find a Solicitor " feature ready. The form is live, the logic is sound. But the reality is that I haven’t onboarded any solicitors yet. This gap: between the technology being built and the ecosystem being populated—has turned into a deep exploration of the invisible architecture required for high-stakes legal tech . Engineering via Subtraction During a period over Christmas, I did some architecture refactory through its first major iteration. Initially, the stack was "standard" for this era: a Next.js frontend, a FastAPI backend, and a RAG (Retrieval Augmented Generation) pipeline fueled by embeddings from 3 legal documents , Vercel cloud . But as I worked, I started questioning the complexity. The models used are already trained on vast amounts of publicly accessible legal da...

It’s been a while for me to refine my thought on this idea, since 2016, till LLM. My first thought, MoDeA (mixture of domain expert and agent), reflected my interpretation of how AI could be embedded into established organisations. But a more bold approach kept tugging at me, not AI inside an organisation, but an organisation built around AI from the ground up. An AI‑first system.   Somewhere along the way, that idea crystallised into something sharper: a multi‑agent proprietary trading system (MATS).   This post is the beginning of what I expect to be a year‑long series - a way to document the evolution of this architecture, and a way to reason about markets without drowning in noise, narratives, and the endless churn of financial content. What started as a desire for clarity has evolved into something far more ambitious: a multi‑agent, AI‑first proprietary trading architecture. Paradigm Selection: Why Agentic, Not Skills The AI industry moves at a speed that makes...