Rahul Sunil

At Certa, I architected Olympus, the shared production platform beneath every new AI agent, reducing time-to-production from months to weeks. It provides multi-tenant isolation, model routing, evaluation, observability, cost governance, and reliable deployment for enterprise AI products.

I took Contracts Agent from a validated client POC to a production distributed Python/FastAPI system in roughly one month, cutting LLM inference cost per contract by 60%. I also led migration analysis from Azure OpenAI to AWS Bedrock, modelling a 10.6× throughput gain for a 100-concurrent-call evaluation pipeline.

For Risk Agent, I built the evaluation and accuracy framework against human-annotated ground truth and delivered changes that raised answer accuracy from 68% to 89% across independently evaluated reports. I re-architected its execution pipeline to reduce request latency by 60% and database connections by 75%, helping compress 10–12 hours of manual analyst review into under 10 minutes.

Before Certa, I delivered search and people-analytics systems at Accubits and led projects spanning UGV telemetry, dark-web OSINT, crop-disease detection, and secure medical records. I'm the sole inventor on a filed workflow context extraction patent and speak about enterprise GenAI and RAG over private data.