Local models
Research in progressExploring model development, adaptation and local inference for Canadian businesses. The aim is useful AI with more choice over where it runs and what it can access.
SHAPEDSI / APPLIED AI
Local models, agents, code and mathematics. Bring a question from your work. Let’s explore how to test it.
Local models Agents & tools Build with code Explore patterns Research notes Choose an object. Explore an idea.
Five research tracks
An imagined place. Work shaped around you.
THE RESEARCH NOTEBOOK
These tracks describe ongoing work. Public model releases and benchmark reports will be identified when available.
Exploring model development, adaptation and local inference for Canadian businesses. The aim is useful AI with more choice over where it runs and what it can access.
Building ways to connect specialised agents, tools and a human reviewer. The work includes task routing, reusable workflows and deciding what an agent is allowed to do.
Exploring coding assistants and the software around them: interfaces, tool connections, testing and review. A useful system needs more than generated code.
Exploring mathematical sequences, structured reasoning and ways to evaluate predictions. Small, understandable experiments help separate an interesting pattern from a reliable method.
Connecting experiments with practical learning for owners and teams. Research notes can explain the question, the method, what happened and what still needs work.
A FEW IDEAS TO INSPECT
These examples run in your browser. They illustrate ideas, rather than the performance of an AI model.
MATHEMATICS / INTERACTIVE EXAMPLE
Choose a sequence and change how many terms you see. The rule is explicit, so you can inspect every value.
F(n) = F(n−1) + F(n−2)
Bar heights show relative values within this sequence.
Deterministic mathematics. This example does not use an LLM.
AGENT SYSTEMS / GUIDED EXAMPLE
Walk through a sample workflow for preparing a project brief. Each step shows its role, before a person decides what happens next.
Start with the task, expected output and boundaries. Step through the example to see the handoffs.
A scripted teaching example. No AI agent is running and nothing is sent or published.
RESEARCH / EVIDENCE BEFORE CLAIMS
A result is easier to assess when the question and conditions are clear. This is the information we want future Lab reports to include.
Our research approachTHE METHOD MATTERS
A demonstration and a measured result are different things. We record the conditions before drawing a conclusion.
Define what needs to happen and how to check the result.
Model, hardware, data, memory and timing in the same record.
Distinguish what is local, external, read or changed.
Explain the method, limits and the questions that come next.
BEFORE WE BEGIN
A founder-led AI research and development lab in Canada, exploring local models, agents, coding systems and mathematics.
The page describes work in progress. Public releases and repeatable benchmark reports will be identified when available.
That depends on the complete setup, including connected tools and services. We agree what runs locally, what connects externally and what each component can access.
Yes. Start with a real task or research question. We can discuss a focused pilot and its review criteria.
SHAPEDSI · Canada
A workflow, a tool or a research idea. Let’s define a useful first experiment.