Simplicial AI // Robust AI systems
From observation to decision and back
Simplicial builds custom AI systems for decisions that matter. They forecast, optimize, and reason, and they can explain how they got there.
Off-the-shelf AI breaks when data is sparse, conditions shift, or someone asks why. We build for those cases, deploy into your workflow, and document the work as applied R&D that can support SR&ED claims.
Every AI system we build runs the same loop. It observes what arrives, represents what is known, reasons about what follows, decides what to do, and learns from the result. Every step can be inspected.
Dynamical systems
Machine learning that forecasts demand, capacity, and risk with stated confidence, and optimization that picks the best next action under your constraints.
Forecasting · Scheduling & resource optimization · Predictive maintenance
Knowledge systems
AI built on language models and knowledge graphs, grounded in your documents and records, that cites the evidence behind every answer.
Document & contract review · Compliance evidence · AI assistants on your data
Explainable by design
- UncertaintyEvery forecast comes with a range, not a single number.
- ProvenanceEvery answer traces back to its sources.
- Decision traceEvery recommendation records the inputs, constraints, and model behind it.
- RobustnessTested against shifted conditions, edge cases, and missing data before it goes live.
See what is changing Decide what to do next
Many operational decisions, like how much to produce, when to service equipment, or where to put capacity, depend on systems that change over time and can only be partly observed.
We combine machine learning that forecasts with a stated uncertainty and optimization that finds the best action under your objectives and constraints. Both update as new data arrives, so plans hold up when conditions shift.
From fragmented information to reasoned decisions
The information behind a decision is usually scattered across documents, spreadsheets, tickets, and the people who know the history. We organize it into explicit entities, relationships, rules, and evidence: a knowledge graph your AI can work from.
On top of it we build AI systems with large language models that answer questions, draft recommendations, and cite the sources behind every conclusion, so your team can check the results instead of taking them on trust.
Why “Simplicial”
Our name comes from simplicial complexes: structures built from points, edges, and triangles. Unlike an ordinary network, they capture relationships among groups, not just pairs, and that's how most real problems are shaped. The figure above is one.
Practical by design Rigorous where it counts
-
Scope
Define the decision
We agree on the decision to improve, the data available, and how success will be measured before any modelling starts.
Problem brief & success metric -
Prototype
Test on your data
We test candidate models on your data, including edge cases and changing conditions, and show what works and what doesn’t.
Working prototype & results -
Deploy
Put it into operation
We integrate the system with your tools and workflow and hand it over with documentation your team can maintain.
Production system -
Improve
Learn from outcomes
We monitor performance against the original metric and update the system as conditions and data change.
Measured impact
Research that ships and supports SR&ED claims
Building AI that holds up in the real world often means answering questions with no standard answer. In Canada, work that resolves technological uncertainty through systematic investigation may be eligible for Scientific Research and Experimental Development (SR&ED) tax incentives.
We already work this way: hypotheses, experiments, recorded results. We keep the technical records as the project runs, so they're ready when you file.
- Technological uncertainties
- Hypotheses & experiment plans
- Experiment logs & results
- Technical narrative
The CRA decides eligibility. We are not tax advisors. We work alongside your accountant or SR&ED specialist and provide the technical side of the claim.
Tell us about the decision and we'll find the next step together
We work with operations, engineering, and product teams. That can be a defined delivery project or longer-term research and development alongside your team.
Tell us which decision you want to improve, what data you have, and what progress would look like. We'll tell you plainly whether we can help.