The challenge
Hexis builds nutrition technology for human performance. It reads an athlete's training and wearable data and tells them what to eat, and when, to fuel the work ahead.
When we started in mid-2024, every athlete had years of history spread across a heavily joined schema, with millions of rows per table. Loading a few days of plans could take minutes, and the company was heading into its seed round.
What we did
Rebuilt the data model
We replaced the joined multi-table schema with co-located JSON storage and rewrote the queries around how the app actually reads data. Multi-day queries dropped from several minutes to under 700ms. Read replicas helped, but the schema change did most of the work, and that turnaround was a key factor in Hexis closing its $2.1M seed round.
Moved to multi-region Google Cloud
We migrated off hosted Supabase to a Google Cloud setup defined entirely in Pulumi. The API runs in Europe, the US and Australia behind a global load balancer that sends each user to the nearest region. PostgreSQL replicates across regions, search runs on a Typesense cluster, and each region has its own Valkey cache. Only the load balancer is public. Databases, caches and backends sit in private networks, and internal tools are behind a VPN.
Took it through SOC 2
Audit logging, access controls and data protection are built into the platform itself, and we drove the company's SOC 2 audit.
Made shipping routine
GitHub Actions builds every change into a container and rolls it out without downtime. The team ships to dev several times a day and to production every week, and a hotfix can go out within the hour. Grafana dashboards and Slack alerts keep an eye on all of it.
Built the product with the team
The backend is TypeScript and Node.js with GraphQL and Prisma, plus a Python engine that does the nutrition math. We connected Apple Health, Whoop and TrainingPeaks, and used Gemini for photo, barcode and voice food logging. Sachin is Hexis's CTO. He leads a team of five engineers and still writes a lot of the code.
We've written up parts of the infrastructure: self-hosting Supabase on GCP, multi-region PostgreSQL, infrastructure as code with Pulumi and locking down access with a VPN and Cloud Armor.
The outcome
Hexis is now trusted by 40% of the professional peloton at the Tour de France and half of the Premier League. In the words of its CEO, the product delivered a 300%+ increase in ARR and found product-market fit.
Tech stack
- TypeScript, Node.js, GraphQL, Prisma
- Python
- React Native
- PostgreSQL, Typesense, Valkey, Kafka
- Google Cloud, Pulumi, Docker, GitHub Actions
- Grafana, Clerk, Chargebee, Gemini