aiQu Labs

ai · END-TO-END CUSTOM AI

One team from sensor and silicon to cloud.

We own the whole stack for your build: data, models, hardware, firmware, deployment and lifecycle support.

Talk to an EngineerPhysical AI & Edge

WHY ONE TEAM

Integration seams are where schedules die.

Splitting an AI product across a data science vendor, a hardware shop and a systems integrator is where schedules go to die. aiQu runs it as one engineering programme, with one plan and one accountable team.

THE PIPELINE

Six stages, each with something you can hold.

Every stage ends in an artefact — a brief, a dataset, a board, a runbook — not a status update.

  1. Step 01

    Discover

    On-site process walk, decision mapping, constraint capture and a written success metric. Ends with a go / no-go.

    • Problem brief
    • Success metric
    • Feasibility call

  2. Step 02

    Data

    Capture rig design, instrumentation gaps closed, labelling standard written, dataset built around real failure modes.

    • Capture rig
    • Labelling guide
    • Dataset v1

  3. Step 03

    Model

    Baseline measured first. Architecture chosen against the latency, memory and power envelope of the target device.

    • Baseline report
    • Model card
    • Eval suite

  4. Step 04

    Edge & Hardware Integration

    Board selection or design, sensor interfacing, firmware, quantisation, thermals, enclosure and on-device profiling.

    • Schematic / BOM
    • Firmware
    • On-board profile

  5. Step 05

    Deploy

    Shadow mode on one line or fleet, then staged rollout. Operator interface, escalation rules and acceptance testing.

    • Pilot rollout
    • Operator UI
    • Acceptance tests

  6. Step 06

    MLOps & Support

    Drift monitoring, scheduled retraining, OTA updates, runbooks and handover training for your own team.

    • Drift monitors
    • OTA pipeline
    • Runbooks

Scroll sideways to see all six stages.

OWNERSHIP

You keep what we build.

The commercial model is deliberately boring: scoped work, delivered artefacts, transferred IP.

  • You own the IP

    Model weights, training code, firmware source, schematics and documentation are delivered to you. No licence that expires if you stop paying.

  • Fixed-scope discovery

    Discovery is priced and time-boxed on its own. If the honest answer is that the problem is not AI-shaped, you get that answer and keep the brief.

  • Documented handover

    Runbooks, retraining instructions and a training session for your engineers. Our support contract should be optional, not structural.

ENGAGEMENT MODELS

Three ways to start.

Pricing is quoted per engagement. TODO: publish indicative ranges once the commercial model is signed off.

  • Discovery Sprint

    2–3 weeks

    Process walk, feasibility assessment, data audit and a costed build plan. Fixed price, ends in a go / no-go.

  • Pilot Build

    8–14 weeks

    One line, one asset class or one fleet. Working system in shadow mode, measured against the baseline.

  • Programme

    Ongoing

    Rollout across sites, with an MLOps and support retainer and a quarterly roadmap review.

NEXT STEP

Start with a fixed-scope discovery sprint.

Two to three weeks, one price, and an honest answer at the end — including no.

Talk to an EngineerExplore Qu