We source GPU capacity when the obvious vendors are full, stand up self-hosted AI stacks that survive provider outages, and back the open-source projects making it all possible.
Services
A small set of things, done carefully.
Capacity is tight and the obvious vendors are full. We help teams find creative paths to the inference they need by helping evaluate lesser-known providers, comparing real throughput, testing performance and quality, and calculating costs against your workloads.
Break out of vendor lock-in and ride out the next major-provider outage. We are experts in self-hosted AI stacks built on open-source tooling, making sure that your AI features keep running on infrastructure you control.
Long-form posts, launch announcements, and tutorials that show your project at its best — written by engineers who actually use the it.
Maintainers should ship features, not babysit Github issues. We help automate the operational toil so the project can keep moving.
New features deserve scrutiny before users find the edge cases. We run targeted validation against real workloads and write up what we find, helping OSS projects ensure their new features are ready for the world to use them.
Move from Fly.io, Heroku, or other small providers onto major cloud providers without breaking what already works. We plan the cutover, handle the data, and stay through the first on-call rotation.
Selected work
Automated the toil of community management for an OSS project by implementing vector search solutions, including a Github bot that linked semantically similar issues and rolling out a Discord bot that answered over 1000 end user questions.
Stress-tested an unreleased feature against representative workloads, surfaced two major issues before launch, and wrote the bug fixes to help the team ship a highly-demanded feature quickly.
Wrote an industry-focused blog series for an open-source tool, including feature launch posts, architecture deep-dives, thought experiment pieces, that drove sustained traffic to their enterprise offering.
Replatformed an early-stage startup’s production workloads from Fly.io to AWS, handling everything from VPC design, ECS Fargate services, database migration, and a deploy pipeline the team could run themselves on day one.
Contact
Tell us about the inference capacity you can’t get, the AI stack you’d rather own end to end, or the open-source project that needs more hands.