Zero-downtime Stripe migration for a multi-site booking platform
Payments / billing Stripe + GoCardless + PayPal
The situation. A live multi-site ticketing & booking platform needed to migrate from a legacy PSP to a modern multi-provider stack. The platform was processing payments in production on the day we started; downtime was not an option.
[Awaiting client case-study details, this block will be populated with: the exact constraints, the dual-write / shadow-write / cutover strategy used, the reconciliation pipeline built, and the measurable outcomes. Currently in drafting, to be published with anonymized details.]
Outcomes (to be confirmed): zero customer-facing downtime across N sites; unified reconciliation between providers; reduced payment failure rate; specific € / ms / hours-reclaimed figures.
Incremental RAG over a codebase for AI agent context and onboarding
AI agents CocoIndex + pgvector + Tree-sitter
The situation. The client was building an internal coding agent that needed real context over a large, evolving codebase. Naive RAG was too slow, too expensive, and went stale on every commit.
[Awaiting case-study details, this block will be populated with: the architecture chosen (CocoIndex + pgvector + Tree-sitter for incremental semantic indexing), the eval harness, the cost/latency improvements, and how it integrated with their agent runtime.]
Outcomes (to be confirmed): cut agent context-build from minutes to seconds; reduced token spend per agent run; enabled onboarding workflows for new engineers.