GroupLabs is a small team of senior engineers and researchers in Calgary & Montreal. Five years of production work for operators in energy, healthcare, robotics, and education.
We are an engineering studio, not an agency. We don’t sell hours or decks — we ship working systems and put our name on them.
In production at
4 institutions · 4 sectors
energy · since 2022
healthcare · since 2024
energy · since 2024
education · since 2023
Selected work
Earlier work, below
Real production deployments — what the system did before, what it does now, and the stack that took it there.
Problem · A heritage forecasting system was missing operational targets at scale and had become impossible to debug across a dozen well sites.
Outcome · Rebuilt the pipeline end-to-end on Ray + Kubernetes. New deployment hits operational SLA across all sites and is now the system of record.
p50 latency
< 12 ms
sustained, 12-month rolling
Stack
PyTorch · Ray · Kubernetes · Prometheus
Problem · A research lab needed a production-quality neural decoding pipeline they could deploy across two clinical sites without losing reproducibility.
Outcome · Built and shipped a JAX-based decoding stack with full reproducibility: every inference is reconstructable from input, weights, and a single seed.
Reproducibility
100%
per-inference, audit-ready
Stack
JAX · numpyro · vendored evals · airgapped deploy
Problem · SMART wanted teachers to generate classroom-ready interactives — games, worksheets, in-lesson activities — without writing code or becoming designers.
Outcome · Built and shipped a teacher-facing generation tool that turns a lesson topic into an interactive game or worksheet ready to drop onto the SMART board, in seconds. Now part of SMART’s interactive classroom suite.
Teacher → interactive
seconds
from a topic line to a playable activity on the board
Stack
TypeScript · React · LLM gateway · activity generators
Earlier work
Full archive at /work
Practice
Production-grade engagements, scoped by phase and quoted by phase. We staff small and senior.
Reliable, high-throughput serving stacks — registries, feature stores, observability, autoscaling. We build them to be operated by your team, not by us.
p50 < 12ms · 99.9% SLO · k8s-native
Tailored models in PyTorch and JAX when off-the-shelf doesn’t fit, supported by reproducible benchmarks and rigorous evaluation. Every result, reconstructable.
PyTorch · JAX · CUDA · vendored evals
Open and internal tools we build along the way: gateways, orchestration, vector search, feature engineering. Each one came out of a real engagement.
Tell · Atmos · Tessera · Norma · Bridge · Mesh
Practice 04 · BuildLess
Detail at /buildless
A working idea, taken to a live test in front of real users in under two weeks, instrumented end-to-end. The sprint ends with a written build-or-kill recommendation backed by behaviour data.
For founders and product teams who would rather spend a week finding out than a quarter committing.
Sprint · 1–2 weeks · from $8k–$15k · larger programs by phase
Test scope
Onboarding redesign
Activation rate vs. existing flow
Sample sprintone of 47
Internal R&D
6 entries · 18-mo trend
Tools and libraries we built to support the consulting practice. Each one started as something a client needed.
| Name | What it does | Ver | Status | |
|---|---|---|---|---|
| Tell | High-throughput LLM gateway · caching · routing | v1.0 | released | |
| Atmos | Multi-cloud ML orchestration | v0.9 | beta | |
| Tessera | Sandboxed code runtime | v0.8 | beta | |
| Norma↗ | Automated feature engineering | v1.0 | released | |
| Bridge↗ | Sub-ms multi-modal vector search | v1.4 | released | |
| Mesh | Distributed compute orchestration | v0.2 | wip |
The studio
GroupLabs is an engineering studio, not an agency. We work directly with founders, product leadership, and engineering leadership — at most one degree of separation from the people who’ll live with what we ship.
We don’t sell hours, decks, or strategy slides. We build production systems and write the runbooks that come with them. The work goes out under our name and we’re available the day after we ship.
We staff small and senior. Engagements are scoped by phase, quoted by phase, and end on a written deliverable. If the fit isn’t there, we say so on the first call.
Address
Calgary, AB · Montreal, QC
› Minimalist line-art outline of Canada (country silhouette only, no province lines), drawn as a single 1px hairline stroke in currentColor so it can be styled. Two small filled dots inside: one at Calgary (~51°N, 114°W — upper-middle of the silhouette, slightly left of center) and one at Montreal (~45°N, 73°W — lower-right). Each dot 4px diameter, in #EA580C (orange-600). Optional thin connecting line between the two dots, dashed (3,3 stroke pattern), at 30% opacity. No labels, no text — the section already labels the cities.
Calgary, Alberta
+1 (587) 700-9968
Montreal, Quebec
+1 (825) 365-9891