White Glove LabsFrançaisBook a call
Work · described by sector

What we have built.

We describe clients by sector and never name them. A number appears only where it was recorded.

Delivered · by sector, never by name

  • Federal government

    Departments wanted generative AI on sensitive information that stays under Canadian control.

    The build
    Built
    A five-layer framework, adversarial testing against MITRE ATLAS, per-department retrieval, and model governance. Canadian cloud and model providers.
    Our part
    Architecture and security. Alexander Leslie led the framework.
    2024 to nowDelivered

  • Small professional-services firm

    The firm wanted agents for routine work without developers.

    The build
    Built
    Our agent platform: a non-technical person describes an agent in conversation and deploys it to email, phone, Slack, webhooks or the web. Five agents on LangGraph, a trace of every run.
    Now
    In production with the pilot firm.
    2023 to 2025Delivered

  • Real-estate developer

    Feasibility studies and investor updates were assembled by hand.

    The build
    Built
    Market research, feasibility modelling, a financial engine (cash flow, NPV, IRR, side-by-side scenarios) and bilingual investor reports drafted on Claude.
    Our part
    Design and build.
    OntarioDelivered

  • Global investment bank

    Risk decisions had to be fast, on the bank’s existing hardware.

    The build
    Built
    GPU-accelerated risk analytics at the edge, a hardware-agnostic control plane that keeps existing infrastructure, and agents for lending and fraud.
    Our part
    The founders led it.
    2024 to 2025Delivered

  • Large Canadian bank

    Customer service and back-office work needed specialised agents.

    The build
    Built
    A bilingual multi-agent platform across cloud and edge: agents for service and compliance, an agent-to-agent protocol, lifecycle governance.
    Our part
    The founders led it.
    2024 to 2025Delivered

  • Marketing agency

    28 specialised AI agents that run a marketing agency’s whole cycle, from campaign strategy to budget.

    2024Delivered

  • Marketing firm

    AI agents inside a Salesforce CRM, with the rules they work under, designed to line up with GDPR.

    United Arab EmiratesDelivered

  • Enterprise AI company

    A multi-agent AI platform, built and delivered remotely from Ottawa.

    United StatesDelivered

Canadian public sector · a one-day exercise

A one-day build for a Canadian public-sector audience

The most directly relevant thing we have built for a Canadian public-sector audience, and the architecture the pilot is designed on. It was one day on open government data, not a client engagement, and the caveats sit beside the figures.

  • Agency 2026 National AI Hackathon

    An agentic system that checks federal grant and contribution spending for compliance, built against a code freeze.

    29 April 2026Ottawa
Challenges
Findings delivered across all ten
Records processed
940, of 1,275,521 available
Investigation flags raised
307, precision not yet measured
AI spend for the day
$299

How it ran, and what it does not show

Pipeline, caveats and the link to the pilot

A national AI hackathon in Ottawa, run with the Government of Alberta, organized by the Alberta Minister of Technology and Innovation and sponsored by Amazon Web Services. Federal and provincial leaders were present. We entered remotely as a two-person team.

The system ran in five stages. Direct SQL ingestion came first, then zero-cost SQL filtering. AI classification followed, round-robined across three cloud providers with automatic failover. The last two stages were reporting per challenge and real-time cost monitoring with threshold alerts.

Source

Figures are from our final submission report for the hackathon, dated 29 April 2026. The detection pipeline ran once. Results were not cross-validated, de-duplicated against legitimate transactions or reviewed for false positives. Token counts were back-calculated from spend against published model pricing. The dollar amounts the system estimated are not published, and no entity it flagged is named. Every finding is a lead requiring investigation, not a conclusion.

We design the pilot on that architecture: multi-provider AI, data lineage, anomaly detection against a learned baseline, and explainable dispositions with an audit trail. The pipeline ran with no human review; the pilot adds human approval before anyone acts on a finding. Security and governance sets out the same control model, and sovereign AI the sovereignty design behind it.

Designs · proposals, not contracts

Systems we have architected and proposed to the Government of Canada

Designs and proposals, not delivered contracts. We describe three of the four at a deliberately reduced level of detail until a decision lands on each.

  • Department of National Defence · IDEaS Competitive Projects

    SAISDF: Multi-Modal AI for Advanced Situational DecisionsProposal

    A sovereign, multi-modal AI framework that fuses data modalities into situational decision support, with compliance enforcement designed into the core rather than layered on top. Submitted June 2026.

  • Department of National Defence · IDEaS Competitive Projects

    Sensor Fusion Compliance EngineProposal

    A modular compliance-by-design engine for multi-sensor fusion. It applies classification, legal and policy rules in real time and produces an explainable disposition and an audit record for every decision. We are prime and design authority, with a Canadian support subcontractor for Protected B hosting and for integration and testing.

  • Department of National Defence · IDEaS Competitive Projects

    Smart Urban Data InsightProposal

    AI middleware that turns existing urban and base sensor feeds into real-time anomaly detection without new hardware. Privacy-compliant handling of civilian data is enforced ahead of analytics, not checked afterward. We are prime and design authority, with a Canadian support subcontractor for Protected B hosting and delivery capacity.

  • Innovative Solutions Canada · Challenge Stream, sponsor National Research Council

    CodeCompass: Deterministic AI-Assisted Compliance CheckingProposal

    Explainable compliance checking for building permit applications. Rather than asking a language model whether a submission complies, CodeCompass compiles the construction code into a machine-readable policy language and runs it over a general rules engine. Every rule returns one of four dispositions: compliant, non-compliant, missing information or uncertain. A working proof-of-concept rule engine exists, from the same compliance-by-design lineage as the defence designs.

We state a proposal’s status where we know it and leave it out where we do not. A design is not a contract, an award or a qualification. The supplier evidence pack lists what we hold and what we do not.

In development · none available yet

Three tools we are building, none of them available yet

Buying professional services by the hour is the easier path for government, because a task-based requirement needs less definition to issue. These tools make the harder, better-value path practical: see what has been bought, reshape a staffing requirement into an outcome, and respond to it accurately.

  • Pursuit IQ

    Turns public contracting history into answers: what was bought, by whom, from whom, on what vehicle and at what price.

    ToolIn development

  • TBIPS to SBIPS Converter

    Turns an hourly staffing need into an outcome: deliverables, milestones, outcome pricing and acceptance criteria.

    ToolIn development

  • Procurement Response Application

    Responds to government solicitations against the rules that govern them; every output traces to its rule. Release 1 targets staffing firms.

    ToolIn development

Next step

Bring one workflow and we will tell you which of these is relevant to it

Source

Dollar figures on this page are the sources’ own or recorded amounts, not our prices. Our prices are in Canadian dollars (CAD), plus GST/HST.

A thirty-minute call scores one candidate workflow and ends with a plain answer about which engagement fits, or that none do yet. The delivery method behind the work above is the three engagements; the control model is on security and governance.

Bring one workflow and we will tell you which of these is relevant to it.

Book a 30-minute callRequest the supplier evidence pack