At Rapid Enterprises, I work across MERN, Next.js, NestJS, and PostgreSQL, taking RAG systems and AI agents beyond demos so they're accurate, observable, and affordable to run.
I rebuilt Grounded from a basic RAG chatbot into an assistant for regulatory and technical documents that gives grounded, cited responses. I added tracing and cost monitoring with Langfuse, using LangChain, Groq, vector embeddings, FastAPI, and Docker.
I created AgentProbe, an evaluation framework that detects when AI agents cheat through trajectory analysis, scaffold analysis, and reward-hacking detection. Testing a Llama 3.3 70B agent showed its success rate dropping from 62% to 50% under my evaluation, revealing that it was gaming the tests.
I also built SafeHer, a women's safety platform with AI danger prediction, live journey tracking, OTP authentication, and an AI legal assistant. I've shipped an AI stock market platform with real-time insights and a multi-agent language learning system as well.
