The questions procurement, legal and IT ask first.

Yes. Every product deploys into an EU or US region of your choice, and Extract and Edge also run fully on-premise. Nothing is used to train shared models. Sub-processors are listed in the DPA, and data-residency commitments are contractual, not best-effort.

We classify each system against the Act's risk tiers, produce the technical documentation and logging the tier requires, and design human oversight into the workflow rather than bolting it on. For GDPR we support DPIA preparation, data-minimisation review, retention controls and a signed DPA. Your legal team is in the room from the first workshop. This is engineering support, not legal advice.

Pilots are fixed-scope and fixed-fee, typically four to eight weeks. You get a working system against one use case, the evaluation harness and datasets, a written production gate agreed up front, and the findings — whether or not the gate passes. Pricing depends on data condition and integration surface; the discovery sprint prices it precisely.

Document workflows built on Extract have gone from pilot start to production in around eleven days when the data path was clean. Assistant deployments across large permissioned corpora more commonly take six to ten weeks. Anything requiring custom hardware follows Xelec's board timelines. We will tell you which bucket you are in during discovery, before you commit.

You own the work product, models trained on your data, prompts, pipelines and documentation. Our products remain licensed to you. There is no clause that lets us reuse your data or derived models for other clients.

No. Our architecture is model-agnostic by default and every deployment records cost and quality per workflow so a swap is an evidence-based decision. Where economics or sovereignty demand it we distil a smaller model you own and run yourself.

Yes. Xelec Edge exists for exactly that: distilled models executing on Xelec-designed hardware at the point of work, offline-first, with store-and-forward sync when a network appears. Extract also has a fully on-premise deployment.

Grounding with enforced citations, refusal when confidence falls below threshold, golden evaluation datasets that gate every prompt and model change, adversarial suites, and drift monitoring after go-live. Where the cost of error is high, we route low-confidence cases to a human by design rather than by exception.

The production gate is agreed in writing before work starts. Pass it and we move to a build partnership or hand off to your team with runbooks; fail it and you keep the findings, datasets and a written recommendation. We would rather lose a project than put a system into production that cannot hold.

Our delivery is oriented around US and European hours, regulation and procurement expectations, which is where most of our work sits. We take engagements elsewhere selectively when the use case fits our products.

Hardware capability most AI vendors do not have: embedded design, sensor integration, board bring-up, thermal and power validation, and manufacturing. When a use case needs inference at the point of work, we design the device and the model together instead of shipping a cloud API and hoping.

Yes. Xelec has run IoT, embedded and electronics product development as standalone engagements for years — firmware, board design, sensor systems, connectivity and manufacturing support. AI is an option on those programmes, not a prerequisite.

You do. Schematics, layout files, bills of materials, firmware source and test fixtures are handed over as part of delivery. Where we reuse an internal library or driver, it is licensed to you perpetually and named in the contract so nothing is hidden.

Yes. We take a design through design-for-manufacture review, pilot build, test-fixture development and certification preparation, then work with your contract manufacturer or ours for volume. Component sourcing and obsolescence are handled as an ongoing sustaining engagement rather than a one-off hand-off.

Book a discovery call. If there is a fit we scope either a two-week discovery sprint or, when the use case is already clear, go straight to a fixed-fee paid pilot.