Context
Cove Neurosciences is a Toronto brain-data startup building software to explore and analyze MEG and EEG recordings. Our principal joined as cloud engineer and became cloud architect — designing the platform end to end over a year-long engagement.
The problem
Neuro signal data is bursty by nature: nothing for hours, then a dense analysis session that needs serious compute. A startup cannot pay for idle clusters, and a startup team cannot spend its week patching servers. The architecture had to make both problems disappear.
What we built
A fully serverless AWS backend: Lambda for API and orchestration, Fargate for longer-running analysis containers, DynamoDB for low-latency state, S3 for signal storage, Amplify carrying the front end, and SageMaker for ML workloads against the signals.
Everything is infrastructure-as-code in CloudFormation, deployed through a GitOps pipeline — with the entire design documented in Confluence and tracked in JIRA so the founding team owns its own platform, not a consultant dependency.
Key decisions
Serverless-first was a cost decision and a headcount decision at once: pay-per-use matches bursty neuroscience workloads, and managed services mean the ops burden rounds to zero.
Documentation as a deliverable: an architecture only the consultant understands is a liability. We wrote it down until it wasn’t.
Outcome
A documented, codified, serverless platform the startup could run and extend without us — which is exactly what a good consulting exit looks like.