Ravinder Sain

At Paytm, I built an autonomous multi-agent SDLC platform where 11 Claude-powered agents take Jira tickets through design, code, review, QA, and human-gated releases across three payment services.

I care about making agent runs decidable, not merely plausible: hybrid retrieval, golden-query evaluations, calibrated LLM-as-judge scoring, per-call cost tracing, credit budgets, and release gates. A measured context change made runs about 4x cheaper, while an eval caught a retrieval regression before it shipped.

I also built a 14-tool MCP server connecting Jira, Bitbucket, SonarQube, Jenkins, and Confluence, and brought a device-health prediction model into production. A second engineering team adopted the SDLC platform, and its first full run reached a reviewed PR in 51 minutes.

Underneath the AI work, I’ve spent four years building Paytm’s payments infrastructure: a double-entry ledger on 14-shard Aurora, commission and fee systems, and Kafka, Redis, and sharded-PostgreSQL services for payment devices. I co-led Paytm’s Indonesia launch as one of two core engineers.