• Developed core search IP for our product -- invented a technique for training strong text-based and image-based embedding models.• Accelerated our synthetic data generation and data processing pipelines by an order of magnitude.• Fine-tuned several 2B and 7B parameter models with techniques like LoRA and QLoRA.• Lead engineer for cutting down search latency by 4x in our online natural language inference classification model.• Lead engineer for productionizing in-house LLMs. Created systems for using 3x larger models with 10x more throughput than what we had before.• Lead engineer for enabling fast inference for vision LLMs (7B parameter models) for offline tasks.

Achieved 40% reduction in model size of summarization models while still matching performance of larger model.

Taught high schoolers AI fundamentals and led them through a facial emotion detection project.

Worked on several real-world projects in which my team and I created and trained computer vision models (on both images and videos), GANs, and multi-arm contextual bandits. Lots of experience with using PyTorch, MXNet/GluonCV, and FastAI to make cutting-edge models (VQ-VAE, YOLOv4, DeepSORT, etc). Gained experience developing frontends using Streamlit, Flask, and NodeJS to present results.