At Forest AI, I'm building Renisa.ai, an AI evaluation and backtesting platform for insurance conversational AI used by insurers including AXA Group and Alteos. I designed and built a framework for testing insurance-agent workflows that automated regression testing and benchmark creation, cutting manual QA effort by ~70%.
I also built a simulated-customer engine with Mastra and ElevenLabs, plus a human-in-the-loop review workflow for approved agent behaviour. A RAG gateway using Pinecone and Langfuse tracing reached ~60 ms p50 retrieval latency with near-100% benchmark success.
At Userology, I rebuilt the campaign-creation UI in Next.js and Tailwind CSS, improving recruitment efficiency by 30%. I refactored application state to Zustand, reducing re-renders by 40%, and fixed session-recording and transcript-alignment bugs that raised LLM-based session analysis accuracy to 95%.
At Infitout Financial Technology Pvt. Ltd., I built 1lyoptions, an options-trading terminal with paper and live trading modes. I also contributed to peerBench, an open, federated platform for benchmarking LLMs, and built an AI comparison playground for multi-model chat and benchmarking.
