At Aria’s Science Ltd., I architected an A2A orchestrator that routes work to specialized agents, with 98% end-to-end success across 500 concurrent sessions. I also built autonomous data-engineering agents that cut manual effort by 90%.
I built an evaluation service that replaced manual QA with regression visibility, benchmarking the agent fleet at 86% exact-pass and 0.86 mean F1. For scientific content extraction, I deployed vision models with 20ms median inference latency and built a retrieval backbone validated across eight BEIR datasets.
At Amazon Development Centre(India)Pvt. Ltd., I developed a FLAVA-based ad moderation classifier that reached 72% precision and 90% recall. Earlier, at Beltech AI Pvt. Ltd., I built a face-recognition system that achieved 99% precision and 95% recall in production.
