Manav Yadav

-Built ERP-integrated AI agents for natural-language report generation via LangChain+RAG+SQL pipelines, cutting manual reporting effort by 70%. -Engineered streaming voice-to-voice interview system (ASR→LLM→TTS)achieving end-to-end latency under 1.5 s. -Fine-tuned open-source LLMs with QLoRA+PEFT on educational corpora; improved answer relevance by 30%. -Built hybrid RAG pipelines (dense + sparse retrieval) over vector databases, boosting retrieval accuracy by 40%. -Reduced LLM inference cost 45% via quantization, batching, caching, and async FastAPI deployment.

-Led NLP/LLM pipeline for financial document intelligence (extraction, classification, summarization); improved throughput 45%. -Designed semantic RAG system with fine-tuned embeddings and vector databases; improved search relevance 35%. -Built REST APIs for CRM automation and intelligent search with FastAPI + Docker; cut manual effort by 40%. -Optimized text preprocessing and batched inference pipelines, reducing latency by 30%. -Implemented real-time multi-object tracking (YOLO, DeepSORT, ByteTrack); improved accuracy 20%.