The problem
Kestrel had a working notebook, a clear thesis, and zero product. They needed to be in front of paying customers in under two months, with an evaluation story tight enough to satisfy technical investors.
What we did
We scoped a single sharp use case, built a RAG pipeline on pgvector, wired in Stripe metered billing, and shipped a custom eval harness that ran on every PR. The team could see — and prove — that quality wasn't regressing as we iterated.
The result
Kestrel shipped to their first paying customer in week 8. The eval suite closed their seed round as much as the product did.