Night Owl IT is the after-hours engineering practice of a Lead Salesforce Engineer with a taste for AI-assisted systems, custom tooling, and home automation built like it has to last.
Night Owl IT is run by Theunis Jacobs, Lead Engineer on a Salesforce team by day, deliberate builder by night. The name is honest: this is what gets made when the day job is finished and the room is quiet.
The day job is enterprise Salesforce: Apex, integrations, AI tooling inside the platform, and the kind of long-running estate work where a sloppy decision costs months. The evenings are where Night Owl IT lives: bespoke systems, agentic AI pipelines, trading infrastructure with proper risk stacks, and home-grade automation that doesn't pretend to be cloud kit.
The common thread is rigour. Threats get named before they're designed around. Plans get written down before code does. Out-of-scope is a decision, not an oversight. The result is a small but unusually considered body of work: single-customer voice assistants with TPM-bound developer consoles, trading bots with seven-layer risk stacks and FCA-aware framing, multi-agent novel pipelines that actually commit to git.
If you need a script in a hurry, this isn't the right shop. If you need someone who'll think about your problem the way an engineer with an enterprise pager rotation thinks about their problems, welcome.
Engagements scope from single tools to multi-month builds. Everything is delivered with documentation, a clear support model, and a written scope you can hand to a finance team.
Senior-level Apex, LWC, Flow, and integration work for mid-market orgs that need engineering depth without the cost of a full hire. Recent work includes Lightning Out 2.0 embeds into external TypeScript apps via OAuth Authorization Code + Frontdoor URLs.
Bespoke pipelines wiring Claude, Claude Code, n8n, and custom C# services into systems that actually do work for you. Multi-agent orchestration, state machines, sliding-window review loops, versioned outputs; built to run, not to demo.
Engineering for retail algorithmic systems, scoped carefully and built FCA-aware. Multi-layer risk stacks, broker-side stop-losses, weekend gap handling, AI-assisted signal validation. Sold as engineering rigour; not financial advice, not signals.
Home Assistant integrations, voice assistants, IoT segmentation, Raspberry Pi satellites, Telegram bots, Docker-hosted brains. For people who want more than off-the-shelf smart home, and who care that it stays local-first.
A small set of principles, applied consistently. They protect the client, the work, and the relationship in that order.
Every system gets a written threat model before code is written. Step-up auth, signed artefacts, network segmentation, audit streams: these are baseline, not bonus.
Scope creep is how good projects fail. Every engagement comes with a written list of things explicitly not being built, and the conditions under which that might change.
Documentation, restore drills, rollback paths, and a written support model are part of every delivery. A system without a recovery story isn't finished.
Personal and exploratory projects that show the way I think about systems. Client work is under NDA.
A multi-room, locally-hosted voice assistant. C# ASP.NET Core brain, Raspberry Pi satellites with wake-word detection, biometric step-up auth via a paired Android app, TPM-bound developer console, signed tool catalogue, Home Assistant integration, segmented network. Productisation plan covers legal, insurance, SLAs, and a per-customer engineering model.
Conservative algorithmic trading bot with seven independent risk layers, AI-validated signal filtering, broker-side stop-losses, daily loss caps, and weekend gap-risk avoidance. Designed against FCA Negative Balance Protection rules for retail clients on OANDA. Audit-ready by intent.
n8n orchestrating Claude Code over SSH to a Windows host. Commander, Author, Continuity, Editor, and Assembler agents collaborate via state files, with sliding-window continuity checks and per-scene git commits. Front and back matter generated, manuscript assembled to docx.
Embedding Salesforce LWCs into a Vercel-hosted Next.js + TypeScript app. Server-side OAuth 2.0 Authorization Code flow, Frontdoor URL exchange, CORS and CSP configuration in the org, and typed React wrappers around the Lightning Out custom elements.
Email is the best way in. I read everything, reply to most things within a couple of days, and only take work I'm genuinely the right person for.