Swapnil Kadekar

I’m an AI/ML Engineer specializing in “Lab-to-Producon,” focusing on getting generative AI from prototypes into real, reliable systems. I’ve worked on end-to-end AI/ML solutions across LLMs, vision, and embedded platforms.

At OStream (an NVIDIA Incepon company), I reported/developed directly with the founder and CEO while tackling diverse AI technologies. I built a private Retrieval Augmented Generaon (RAG) system (vector database + DeepSeek R1) for a $7B rail/transportaon project, addressing high-speed correspondence from pilot to producon.

I’ve fine-tuned and deployed models for multimodal and NLP use cases, including VLMs and transformer-based pipelines (BERT, BART, sentence transformers) for tasks like sentiment analysis, classification, clustering, and entity recognition. I also led performance benchmarking of embedded large language and vision models on NVIDIA Orin and Rockchip NPU devices for real-time inference optimization.

I bring strong engineering rigor, including building production-grade embedded vision transformer camera systems using TensorFlow, creating reliable data pipelines with Databricks, and developing frameworks like NVIDIA Lab’s generative AI API gateway (Gapi). I’m especially driven to ship practical, high-quality AI that works under real constraints.