Blockchain wallet backend

The problem

The wallet's backend talked to blockchain nodes directly for nearly everything, including hot-path reads like account balances. That made it slow and fragile: p95 balance reads sat around 600ms, and every user action piled more load onto the nodes.

What I built

I designed and implemented a new microservice backend for the wallet ecosystem, with a read-through caching layer in front of every read-from-node operation - balances and everything else the wallet pulled from chain. The results:

• Direct node requests dropped by about 60%.
• p95 balance-read latency fell from ~600ms to ~150ms.

I also extended the Cosmos SDK chain (Go) with 2 custom modules the standard SDK didn't cover: an on-chain exchange, and a module that converted the chain's native token into AI-usage tokens - the credits users spent to run AI models. Both went from zero to production alongside the wallet.

Leading the team

I led a team of 3 engineers and put the fundamentals in place: code review and CI/CD practices that cut the release cycle from ~2 weeks to ~3 days while raising code quality.

A little AI where it paid off

I integrated an AI assistant for common blockchain operations, which removed about 30% of routine support requests and shortened the path from new user to first transaction.