Work  ·  Bottleneck → impact

The shapes of systems we ship. No client names.

No logos wall, no case-study deck. Most of the work is internal, sensitive, or under NDA — and the rest is our clients’ to talk about, not ours. So we publish what we can: each pattern as we scope it — bottleneck, how it’s financed, the bounded sovereign solution, measurable impact. Sector and revenue band only. If one looks like a problem on your desk, that’s the conversation to start.

68
Weeks, typical build engagement
94%
On budget, on date
100%
Engagements include literacy & handoff
30d
Post-launch support, every project
structured-extraction · illustrative Queue Review Audit
intake_2841 · standard · 12 pages extracted 87s · 38/38 fields
SubmitterField 010.99
Amount$2,840.000.98
Term12 months0.99
Effective2026-08-010.99
Variationindexed +1.5%0.94
Edge clauseSee section 14(c)human
Add-ons1 unit · included0.96
Hold-back$2,840 deferredhuman
SHAPE 01   Structured extraction  ·  4–6 weeks  ·  professional services  ·  ~$10–50M revenue

Pull structured data out of long, messy documents.

The bottleneck

A regulated intake team re-keying 38 fields out of 12-page contracts by hand — multi-hour reviews, fatigue errors late in the day, and a backlog that grew faster than headcount could clear it.

Federal & financial alignment

Scoped as a bounded productivity asset, so the build may be eligible for accelerated expensing under the Productivity Super-Deduction (subject to current program terms).

Structured for the Productivity Super-Deduction
The sovereign solution

A bounded, Level 1–2 extractor that reads the document, scores its own confidence, and routes low-confidence edge cases to a human checkpoint. Read-only by default — it drafts, never writes back without a human key. Architected for Canadian data residency and evaluated against the client’s real historical documents before production.

The measurable impact illustrative

On the illustrative evaluation set below: per-document review time falls from hours to roughly 90 seconds, field-level accuracy lands in the low-to-mid 90s, and only the edge cases reach a person — with the corpus kept onshore the whole time.

Field accuracy · illustrative
93–96%
Time / doc · illustrative
~90s
Routed to human · illustrative
5–12%
Eval set · illustrative
1.2k

Figures are illustrative of the pattern, not drawn from a single named engagement.

Bounded · Level 1–2 Human-in-the-loop Canadian data residency Eval harness

SHAPE 02

Classification & routing for high-volume intake.

Bottleneck. A logistics operator hand-triaging thousands of calls, forms, and emails into the wrong queues.
Solution. A bounded Level 1–2 classifier mapping intake to a fixed taxonomy with a confidence floor; below it, routes to a human. Read-only, Canadian-resident.
Delivered so the client could pursue COJG subsidy
Build  ·  2–4 wks  ·  logistics  ·  ~$5–20M rev
SHAPE 03

Internal Q&A grounded in your real documents.

Bottleneck. A professional-services firm where senior staff fielded the same policy questions all week — answers buried across folders, tickets, and heads.
Solution. A bounded, read-only assistant that answers only from the client’s own corpus, with citations and a confidence floor; outside it, hands off rather than guesses. Level 1–2, Canadian-resident.
Delivered so the client could pursue COJG subsidy
Build  ·  3–6 wks  ·  professional services  ·  ~$10–50M rev
SHAPE 04

Workflow automation across the tools you already pay for.

Bottleneck. A wholesale-distribution back office shuttling data between email, docs, a CRM, Slack, and the spreadsheet nobody admits is load-bearing.
Solution. Bounded n8n / Make / code workflows, LLM steps only where they earn it, blocked from external writes without a human gate — plus a kill switch and audit log. Level 1–2, Canadian-resident.
Structured for the Productivity Super-Deduction
Automate  ·  2–4 wks  ·  wholesale distribution  ·  ~$20–100M rev
SHAPE 05

AI-ready digital foundations — sites and product surfaces.

Bottleneck. A services firm whose site and internal tools couldn’t host the bounded automations they wanted next — everything bolted on after launch.
Solution. Sovereign-hosted surfaces built to target WCAG 2.1 AA, architected to integrate bounded, human-in-the-loop automations later — real components, real states.
Web  ·  3–6 wks  ·  professional services  ·  ~$5–20M rev
SHAPE 06

Defensive AI literacy & upskilling on your real work.

Bottleneck. A finance-sector team using consumer AI tools ad hoc — no shared rules, real exposure to leaking data into prompts they didn’t control.
Solution. Half-day to multi-week workshops on the client’s own tools and the workflow we shipped — defensive, PIPEDA-safe prompting, practiced on real artifacts, not toy prompts.
Delivered so the client could pursue COJG subsidy
Enable  ·  half-day to 6 wks  ·  finance  ·  ~$50–250M rev
Same architecture, every shape

Every pattern runs on the Sovereign Security Stack.

The four parts above aren’t bespoke per project — they sit on one bounded architecture. It’s what lets every shape stay Level 1–2, read-only by default, and Canadian-resident without a custom security story each time.

See the full Sovereign Security Stack

Layer 01

Sovereign Compute (Data Residency)

Architected for Canadian data residency.

Layer 02

Bounded Autonomy (Read-Only by Default)

Level 1–2, blocked from writing without a human key.

Layer 03

Human-in-the-Loop (The Approval Gate)

The AI proposes; your team executes.

Layer 04

Recoverability (The Kill Switch)

Immutable audit logs and an instant revert to manual.

Why no logos wall

A note on discretion.

Most systems we’ve shipped sit inside operations, presales, legal, and support stacks — places a company doesn’t advertise because the competitive edge lives there. We’d rather have the long, referral-driven relationship than the screenshot.

What’s here is what we’re free to publish: the shape of the work, how it’s financed, and the rules we run it by. We’d rather show you a real (NDA’d) system live on a call than post a sanitized screenshot — facelessness is the point, the demo is the proof.

Book a screening call — we’ll run one live

Yours next

If one of these resembles your problem, let’s talk.

A thirty-minute call. We read your brief before we meet. If it’s not a good fit, we’ll point you toward someone better — and probably name them.

Map Your AI Funding & Compliance Strategy All services
Pick a track

Recognized your bottleneck above? Route into the right track: on the growth track, see how these patterns get funded — Growth & Funding . On the compliance track, see how they stay bounded and AIDA-ready — Risk & Compliance .