StarApple AICanada

Detail

What each practice delivers.

Each entry states why organisations start there, everything you receive, and how long it runs. The durations are ranges from delivered engagements. Where one practice depends on another, that dependency is listed in its specification.

CA·01

AI audits

Aurora, AI audit and readiness assessment

An inventory of every model and AI feature running in your organisation, each with a risk rating and a costed plan.

Duration4 to 6 weeks
Who it is forRisk committees, CIOs, and boards about to approve AI spend
Ends withA model register and a prioritised 90-day roadmap
InstrumentsAlgorithmic Impact Assessment, OSFI E-23, ISO/IEC 42001, PIPEDA, Law 25

Why organisations start here

AI arrives in an organisation through several doors at once: a licensed assistant, a "smart" feature switched on inside a vendor product, a spreadsheet macro somebody wrote, a team paying for its own tooling. An audit produces the register of what is running, what data feeds it, and which decisions it already touches, which is the input every governance and strategy decision afterwards depends on.

What you receive

  • Model and tool register covering sanctioned, vendor-embedded and unsanctioned AI
  • Data landscape audit: where data lives, its quality, lineage and accessibility
  • Readiness assessment across every business function
  • Opportunity matrix: what to automate, augment, or deliberately leave alone
  • Risk and compliance gap analysis against the Canadian instruments above
  • Algorithmic Impact Assessment drafted for your two highest-risk systems
  • Prioritised 90-day roadmap with cost, owner and expected return per item
  • Executive briefing delivered live to the board or risk committee

CA·02

Team AI training

Lucid, applied AI training for teams

Every participant leaves with a working assistant for their own role, a prompt library the team keeps, and a usage policy your legal group can adopt.

DurationTwo 90-minute sessions, one week apart
Cohort sizeUp to 25, hands-on throughout
Who it is forWhole departments: operations, finance, legal, marketing, service
Ends withA working assistant per person and a drafted usage policy

Why organisations start here

Untrained AI use produces inconsistent output, confidentiality exposure, and fluent answers that nobody checks. Prohibition moves the same activity to personal accounts, where there is no audit trail and no way to see what was pasted in. Training replaces the prohibition with competence and a rule set specific enough to follow.

What you receive

  • Session one covers foundations: how models work, prompt anatomy, privacy
  • Session two covers advanced work: multi-step workflows, chain-of-thought, quality control
  • A role-specific AI assistant built by each participant during the session
  • Five or more workflow templates per role, written in the room
  • Guardrail training on what must never enter a general-purpose tool
  • A shared prompt library your team keeps and extends
  • A drafted AI usage policy ready for HR and legal to adopt

CA·03

Machine learning training

Boreal, machine learning practitioner programme

Analysts who already know SQL and Python finish having built, validated and deployed a model on your own data.

Duration8 weeks applied, or 12 weeks full track
Who it is forAnalysts, actuaries, engineers and BI teams with SQL and Python basics
FormatWeekly live lab plus a capstone on your own data
Ends withA deployed model built by your team, on your infrastructure

Why organisations start here

Canada trains machine learning researchers at Mila, the Vector Institute and Amii, then watches mid-sized firms bid for them against the banks. The faster route into a mid-sized organisation runs through the people already inside it: the analysts who understand your data are a structured term of practice away from building production models on it.

What you receive

  • Supervised learning: regression, classification, and honest evaluation
  • Feature engineering on your own warehouse rather than a teaching dataset
  • Time series and demand forecasting for operational planning
  • Model validation, drift detection, and the documentation E-23 expects
  • MLOps fundamentals: versioning, reproducibility, deployment, monitoring
  • Fairness and bias testing as a standard step of the workflow
  • A capstone model, reviewed by us and deployed into your environment
  • Individual certification and a written capability assessment per participant

CA·04

AI model development

Atlas, bespoke AI and model development

A model trained on your own history, integrated into the systems your staff already use, documented to the standard internal audit expects.

DurationScoped per engagement, 3 to 9 months
Who it is forOrganisations with proprietary data and a decision worth automating
Data residencyCanadian-region deployment where residency is required
Ends withA production system, monitoring, and full knowledge transfer

Why organisations start here

A licensed general model answers questions any firm in your sector could ask. The decisions that move your margin depend on patterns held in your own history: which claim to fast-track, which member lapses next quarter, which shipment will miss its window. A bespoke model reads those patterns, and it is yours to retrain afterwards.

What you receive

  • Problem scoping and a written feasibility verdict before any build begins
  • Data preparation, feature engineering and leakage review
  • Model development, training and validation against a held-out baseline
  • Retrieval and agent architecture where a language model is the right tool
  • Integration into the systems your staff already use: CRM, core, warehouse
  • Monitoring dashboard tracking drift, performance and cost
  • Model documentation written to satisfy internal audit and OSFI E-23
  • Knowledge transfer so your team can retrain it without us

CA·05

AI safety and alignment

Sentinel, AI safety and alignment practice

Adversarial testing against your deployed system, with an evaluation suite your team runs on every release and a stated pass mark.

Duration6 to 10 weeks, then retained review per release
Who it is forAnyone putting a generative system in front of customers or caseworkers
Ends withAn evaluation harness your team runs in CI
InstrumentsVoluntary Code for advanced generative AI, Algorithmic Impact Assessment, ISO/IEC 42001

Why organisations start here

A system tested the way it is demonstrated gets a handful of cooperative prompts, all of which work. The population that arrives after launch is larger, less cooperative, and distributed differently from anything in the demo. This practice treats safety as engineering work with artefacts: an attack log, a threshold, a test suite that fails the build.

What you receive

  • Red-team exercise against your deployed system, with a written attack log
  • Evaluation harness: automated test suites you keep and run in CI
  • Hallucination and grounding measurement with a stated acceptance threshold
  • Bias and differential-performance testing across protected characteristics
  • Prompt-injection and data-exfiltration testing for agentic and retrieval systems
  • Refusal and escalation design: what the system declines, and who it hands to
  • Human-oversight design meeting the Algorithmic Impact Assessment's intervention requirements
  • Incident runbook: detection, containment, notification, post-mortem

CA·06

AI strategy

Meridian, AI strategy and investment roadmap

A costed two to three year roadmap naming which AI investments get funded, which get closed, and the conditions under which each judgement changes.

Duration6 to 8 weeks
Who it is forExecutive teams and boards setting a two to three year direction
Pairs withAn audit first. We decline strategy work on an unmapped estate
Ends withA costed, sequenced roadmap your CFO can defend

Why organisations start here

Pilot sprawl is the failure mode we see across Canadian mid-market and public-sector clients: a dozen promising experiments, none of them resourced to production, all of them defunded together when the budget tightens. This practice forces the allocation question, including the part where specific experiments get closed and the money moves.

What you receive

  • Competitive and sector scan of what is actually deployed in your Canadian market
  • Value-pool analysis mapping AI opportunity to your own P&L lines
  • Build, buy or partner decision for each opportunity, with the reasoning shown
  • Two to three year sequenced roadmap with funding gates and kill criteria
  • Operating-model design: where AI capability sits and who owns it
  • Talent plan (hire, train internally, or partner) with cost per route
  • Board-ready narrative and the investment case behind it

CA·07

AI policy and governance

Charter, AI policy and governance development

A governance framework built on ISO/IEC 42001 and wired into the release process you already run.

Duration8 to 12 weeks
Who it is forRegulated firms, public bodies, Crown corporations, and their counsel
Ends withAn adopted framework, a live committee, and trained reviewers
InstrumentsISO/IEC 42001, TBS Directive and AIA, OSFI E-23 and B-13, PIPEDA, Law 25

Why organisations start here

Governance takes effect at the point where it sits on the path a system already travels to production: a gate in the release process, a named approver, an intake form short enough that an engineer completes it rather than routing around it. We write the framework and then install it in that path, and we train the people who will staff the reviews.

What you receive

  • AI governance framework structured on ISO/IEC 42001
  • Risk-tiering methodology aligned to the AIA and federal high-impact logic
  • Acceptable-use, procurement and vendor-diligence policies
  • Model lifecycle standard: approval, validation, monitoring, retirement
  • AI committee terms of reference, RACI, and meeting cadence
  • Intake and review gates embedded in your existing change process
  • Reviewer training so the committee can assess submissions without us
  • Regulator-ready evidence pack and a twelve-month maturity plan

Next step

If none of these is obviously the one.

Describe the decision in front of you and we will name the practice that fits, or tell you that none of them does yet.

Book a scoping call engage@starapple.ai