Tushar Vatsa Email & Phone Number

San Jose, Ca, Us

San Jose, Ca, Us

- Contextual Recommendations using Large Language Models * Training 7B Models on Express Assets for next asset prediction. * Built Automated Evaluation using vllm inference. * Built a demo using gradio for manual analysis of failures. * Built an Asynchronous Sensei CPF Service which can handle multiple concurrent requests. * Built the script for data cleaning of Assets Markup Language. * Built the Monitoring playground on wandb for training and evaluation analysis.- IntentToAction * Based on user's prompt on Express, built an AI engine which can match those prompts to Actual Actions. * Trained the Orchestrator using synthetic data, for 7B and 2B Models. * Benchmarked the performance of different models for Orchestration. * Built the gradio demo for analysis of failure cases. * Currently working on building the RAG for Adobe Search's Knowledge graph which can add more context for Orchestration.- AdobeOne Models(Patent : Contrastive Multilingual Language and Image Pretraining and Finetuning) * Evaluation of AdobeOne English and MultiLingual Models for stratified behavioral data. * Built the gradio demo for manual analysis of results and failures for retrieval. * Service deployment for text and image queries returning AdobeOne Embeddings.

•Set up the data pipeline from the Amazon S3 buckets to the local server for efficient model training in a docker container.•Automated the quality assurance script to detect errors in the gold labels provided by annotators which reduced the overall pre-processing time significantly.

Master'S Degree, Electrical And Computer Engineering

B.Tech, Electronics And Instrumentation

Education record