Technology · 05
Boring choices, made deliberately.
Every tool here is one we would still choose in three years, when someone else is maintaining it. Novelty is a cost your business pays, not a feature we get to enjoy — see how it fits into our delivery process.
Selection criteria
- Can we hire for it?
- If your next developer cannot be found in a week, the technology is a liability regardless of how elegant it is.
- Does it fail loudly?
- We prefer tools that break at build time over tools that break at 2am in production.
- What does it cost at 100×?
- Pricing that is free at launch and ruinous at scale is a decision deferred, not avoided. We model it up front.
- Can we leave?
- Every layer has an exit. Your data is portable, your models are swappable, and nothing important is hostage to one vendor.
The intelligence layer
Model-agnostic,
on purpose.
We build against an abstraction, not a vendor. When a better or cheaper model ships — and one always does — your system moves to it without a rewrite. Lynq is this layer, productised.
- Reasoning & generation
- Claude, GPT, and Gemini, selected per task on cost, latency, and quality — not loyalty.
- Voice
- Realtime speech agents that answer calls, qualify, and book — with a clean handover to a human when it matters.
- Retrieval
- Answers grounded in your inventory, your documents, and your policies — with citations, so a wrong answer is traceable.
- Evaluation
- Every AI feature ships with a test set. If we cannot measure it, we do not put it in front of your customers.
Engineering standards
The budget every build ships against.
95
Lighthouse, all categories
2.5
Largest contentful paint
0.1
Cumulative layout shift
AA
WCAG 2.2 conformance
This is the budget every system in our selected work ships against — not a claim, a standard.