At NKS Technologies, I architected a production RAG system with LangChain, LlamaIndex, FAISS, GPT-4, and Pinecone. It became the company’s primary client-facing AI product within a quarter of launch and cut support ticket volume by 34%.
I fine-tuned Llama 2 and evaluated Llama 3 variants on proprietary documents, and built a real-time voice AI pipeline using Whisper, Mistral, and ElevenLabs. The voice pipeline served 500 concurrent sessions with end-to-end latency under 1.2 seconds.
At Virtual Force, I delivered a legal document classification engine that achieved 94% F1 across 12 categories and processed 50,000 documents per day. I also built an anomaly detection system for a manufacturing client using industrial sensor data.
Earlier, at EffectiveSoft, I engineered data pipelines and deployed a churn prediction model that contributed to a 17% reduction in quarterly churn for the flagship SaaS account. At Emaar Properties, I built APIs and responsive web interfaces, and worked on integrations including OpenAI-powered features.
