Honest answers to the questions small and mid-size teams in the Lower Mainland keep asking before they bring me in. If yours isn’t here, drop it in the form below — I reply the same day.
I’m based in the Lower Mainland, British Columbia, Canada. Most client work happens in person within an hour or so of home — Surrey, Vancouver, Burnaby, Richmond, Delta, Langley, New Westminster, and Coquitlam — with remote engagements running anywhere in the world for organizations that just need the engineering brain on the problem.
Start with the free AI readiness audit: a no-cost 60-minute review and a short written report on where private AI (and tighter IT generally) would actually save you money. Real CAPEX-vs-subscription math for your situation, what’s safe to keep in-house vs. cloud given your data, and one quick win you can act on immediately. Yours to keep, no obligation.
The free audit is free. The Private AI Starter pilot is a fixed-scope project quoted after a brief scoping conversation — one GPU server, one or two use cases, IdP-backed login, handover runbook. The Managed IT + AI Concierge retainer is a monthly fee sized to your stack. Bespoke work is hourly or fixed-bid by scope. Whatever the shape, the number is the number — no surprise line-items.
For private AI engagements — no. That’s the entire point of the offering. Models run on hardware you own, prompts and uploaded files stay on your LAN, and you keep the audit log. I configure the stack, hand over runbooks, and operate only what we’ve explicitly contracted.
For moderate-to-heavy team usage, a single capable GPU server amortized over 2–3 years often beats per-seat AI subscriptions on a 36-month total cost basis. Exact crossover depends on your token volume. You also get data sovereignty, customization (RAG over your own docs, custom prompts and tools), full auditability, and no forced model deprecation. Trade-offs I’m transparent about: hardware capital cost upfront, ops responsibility (which is what I provide as a service), and slightly more friction to upgrade to the latest frontier model. For most business tasks, mid-range open-weight models are sufficient.
Mostly 7B–14B parameter models from the Llama, Qwen, and Mistral families — sufficient for most business tasks and they run comfortably on a single mid-range GPU. For workflows that genuinely need frontier capability, I’ll wire in an API model (Claude, GPT-4-class) with explicit guardrails so you know exactly what leaves the network and when.
Yes — deeply. My day job is Windows Server + SharePoint + M365 administration. Strong on Entra ID, conditional access, MFA rollouts, AD cleanup, hybrid identity, and Tier-0 separation for admin accounts. If your shop is mid-migration from on-prem AD to cloud-only Entra (or living comfortably in hybrid), I’ve been there.
Yes to all three. Primary daily driver is a MacBook Pro M2; deep experience with Windows Server + Microsoft 365 (day job), and Linux (home lab, most consulting work, every client GPU server I’ve racked).
The clearest demos are the two AI tools running on this same infrastructure: LLM Chat (private on-prem Ollama, nothing leaves the LAN) and Tower (browser SSH + AI assistant for ops work). Both are tools I use every day — not marketing demos. Client work itself is covered by NDA, but I can talk through patterns and architecture in an audit.
Yes — availability is signalled by the Available for new engagements badge on the home page. Audits are always open even when project capacity is tight.
PHP, Python, JavaScript/TypeScript, Bash, PowerShell, and a fair bit of SQL on the code side. nginx, Apache, systemd, Docker, and the usual Linux sysadmin tooling on the infrastructure side — whatever the job calls for.
Drop your question (or a quick description of what you're trying to figure out) below. I reply the same day, no sales funnel, no SDR — straight to me.
Prefer email? info@sd-techsolutions.com · Same-day reply.