Mandava Kashyapsai

I’m an AI/ML Engineer building production-scale Generative AI solutions centered on LLM pipelines, RAG, and real-time Voice AI. At Vocab AI, I automated customer call QA for enterprise clients (CRED, Lenovo, Libas), processing 4,000+ calls/day across 35–40 evaluation parameters.

I’ve driven measurable cost and performance improvements by replacing Gemini-based evaluation with self-hosted Qwen2.5-7B (llama.cpp), cutting monthly AI costs from INR 100K+ to INR 60K while maintaining throughput. I also built streaming conversational pipelines (ASR → LLM → TTS) for outbound voice campaigns, supporting 1,000+ calls/day with sub-2s response latency.

I enjoy engineering self-hosted inference systems and scaling them via batching, Kafka, and Nginx load balancing, with optimized GPU utilization through containerized Docker deployments. My work also includes a LangGraph/RAG banking Voice AI assistant (90% intent resolution, sub-2s), and I earned an IEEE Best Paper Award for real-time MRI synthesis research—so I bring both applied engineering and rigorous problem-solving to every rollout.