SIF — Strategic Informatics Facilitation
Purpose-built AI models for industries where data cannot leave
Compact, task-specific models for accounting, banking, taxation, healthcare and other regulated domains — embeddable in the platforms you already run, installable on your premises, air-gapped when required. No chatbots. No generative AI.
- Ottawa
- Based in Ontario, serving Canadian businesses
- Non-generative
- Models that classify, extract and match — not chat
- Air-gapped
- On-premises deployment, in development
What we build
From niche use case to installed model
Four pieces of one idea: find the narrow task, distill a compact model that holds it, package it so it embeds and installs where the work already happens — and prove it against a baseline before anything depends on it.
- Purpose-built, not general-purpose
Domain AI Models
Compact, task-specific models for accounting, banking, taxation, healthcare and other regulated domains — built to classify, extract and match, never to chat.
How it works - How the models get small
Teacher–Student Distillation
Large teacher models supervise the training of compact student models for one niche task. Only the student ships — small enough to run on hardware you own.
How it works - A solution, not a science project
Embeddable & Installable
Models packaged as software: embeddable in the platforms you already run, installable on your premises, fully air-gapped where the data demands it.
How it works - Where an engagement starts today
Use-Case Discovery & Pilots
Working with organizations to find the narrow, high-volume tasks a compact model can genuinely absorb — and piloting against a measurable baseline.
How it works
Private AI: your data never leaves your office
Financial, health and legal records are exactly the kind of data that cannot go to a public cloud model. SIF is developing compact, task-specific models — distilled from teacher models, trained on vetted niche domain data — designed to run locally on client premises.
This is a research and development program, not a product on a price list. We describe it here because it is the direction of the firm — and because you should know what is shipping today and what is still being built.
Read about the program-
Data stays on site
The design goal is a model that runs on hardware in your office, so client financial records never travel to a third-party cloud service to be processed.
-
Vetted domain data
Trained on curated, checked domain material rather than the open web — narrow scope is what makes a small model viable.
-
Small by design
Teacher models supervise training; only the compact student ships, sized for ordinary on-premises hardware with no external AI dependency at run time.
How we work
Three commitments
- 01
Task models, not chatbots
Nothing we build generates content or holds a conversation. Every model does one narrow, checkable job — which is what makes it small, auditable and deployable where data cannot travel.
- 02
A person stays in the loop
Models remove the re-typing and the classification grind. Judgement, review and sign-off stay with people.
- 03
We say what is real
Work in progress is labelled as work in progress. You will not find invented case studies or borrowed metrics on this site.
Tell us what the work actually looks like.
A first conversation costs nothing and is usually enough to tell whether we are the right fit. If we are not, we will say so.