Devansh Agarwal Email & Phone Number

New York, Ny, Us

Lawrence Township, Nj, Us

Lawrence Township, Nj, Us

• Trained and fine-tuned billion-parameter models from scratch, advancing the capabilities of large-scale AI systems.• Collaborated closely with stakeholders to incorporate feedback and create measurable impact in AI-driven projects.• Developed LLM-based agentic applications to accelerate clinical trials, improving efficiency in the drug development process.• Designed and implemented an automated tumor assessment pipeline using supervised and self-supervised learning algorithms.• Architected and optimized a cloud-based computing platform on AWS, enhancing data science team productivity.• Created an NLP-based pipeline for extracting key attributes from FDA drug approval documents, providing valuable insights into regulatory processes.

Morgantown, West Virginia, Us

• Developed and implemented ETL pipelines to analyze petabytes of data in search of faint astronomical signals. The pipelines were built using Airflow, Python, and CUDA-based GPU accelerated codes, and they were able to search for faint astronomical signals 20 times faster than realtime.• Implemented extensive monitoring systems with Elasticsearch for geospatial data. The monitoring systems provided real-time insights into the performance of the pipelines, and they helped to identify and resolve any issues that arose.• Created Slack bots for easy user interaction with the pipeline. The Slack bots allowed users to interact with the pipeline in a natural way, and they made it easy for users to get the information they needed.

Morgantown, West Virginia, Us

• Led a team of 22 astronomers worldwide and conducted a 300-hour survey with Australian Square Kilometre Array Pathfinder Telescope.• Processed 100+ TB data in real-time to search for faint astronomical signals and discovered FRB 180417.

Morgantown, West Virginia, Us

• Wrote ETL pipelines using Python, Elasticseach, InfluxDB, and RabbitMQ with CUDA-based GPU accelerated codes to search for faint astronomical signals in realtime.• Deployed the pipeline at two of the world's largest telescopes and searched more than 400+ TB data in realtime.

Morgantown, West Virginia, Us

• We created 11 multi-input CNNs with Tensorflow, each with recall and precision above 99.5%.• Deployed the models at 3 of the world's largest telescopes.• Create a python package and open-sourced the models which discovered 100+ real astronomical signals.

San Francisco, Ca, Us

• Created DocClean, a web app to clean up images with crumpled backgrounds and stains.• Enhanced Google Tessaract’s accuracy up to 7% by removing dirty backgrounds.• Utilized AWS EC2 instances for training and git for system control and code reviews.• Leveraged Tensorflow to implement and train Autoencoders and Cycle-GAN based models.

• Piloted gravitational waves analysis and developed a novel unified framework to search.• Developed a python-based framework for running large scale Monte Carlo Simulations.• Presented research findings at the Astronomical Society of India conference with 300 participants.

Doctor Of Philosophy (Phd), Physics

Integrated Bs-Ms, Physics