Saksham Doshi

I am an AI/ML engineer specializing in retrieval-augmented generation, conversational AI architectures, and real-time computer vision systems. I combine practical deployment experience with model optimization to build efficient, low-latency inference pipelines.

At Metadee AI I designed and deployed a RAG pipeline using sentence transformers and LLMs on AWS EC2, cutting latency by 35% and improving BLEU by 22%. I also engineered dense embedding and vector search infrastructure to enable contextual response retrieval.

My projects span end-to-end conversational agents—integrating speech-to-text, neural TTS, and LLM reasoning—and a real-time yoga pose detection system using Mediapipe and OpenCV with live inference feedback. I focus on balancing cloud and edge trade-offs for predictive local response generation.

I hold a B.Tech. in Computer Science and completed a minor at IIT Guwahati, with research accepted at ICIECT on a latency-aware local inference framework. I seek roles where I can apply scalable ML deployment skills and optimize multimodal, real-time systems.