Astarag Mohapatra

About Astarag Mohapatra

In 2018, I stumbled upon a YouTube video by DeepMind on an artificially intelligent agent named AlphaGO. Hitherto, no artificial agent was built to beat humans in GO, but this AI could beat the world champion Lee Seedol by a considerable margin. AlphaGO came up with a novel move in the 37th turn, which redefined the history of this game and it helped humans to go beyond limits. So I was drawn to this symbiotic relationship between humans and machines and it was the start of my journey in the field of AI and ML. Since then I have been involved in multiple projects, research and work experiences to learn the necessary tools and skills to fulfil my ambitions. Having a high-level picture view that the relationship with AI will be conducive for us, will inspire me to put in efforts every day.As a Computational Science student at Indiana University Bloomington, I am passionate about exploring the symbiotic relationship between humans and machines and applying artificial intelligence and machine learning to solve real-world problems. I have a strong background in mechanical engineering and electronics and communication engineering, which gives me a solid foundation in mathematics, physics, and programming. I have gained valuable hands-on experience in various domains of AI and ML, such as deep learning, reinforcement learning, natural language processing, computer vision, and finance. I have worked as an Advanced Analytics Intern at BeiGene, where I developed and deployed deep learning models for image segmentation and classification of medical images. I have also worked as a Graduate Research Assistant at Indiana University Bloomington, where I implemented and compared deep reinforcement learning algorithms for neurogym environments, and analyzed the neural signals of artificial agents and animal brains. Additionally, I have contributed to the FinRL and FinRL-Meta projects at AI4Finance, where I wrote blog posts, paper explanations, trading demos, and bug fixes for the FinRL library, and led the development of end-to-end hyperparameter optimization and explainability of DRL algorithms in the financial world. I have acquired multiple certifications in deep learning, sequence models, and neural networks from Coursera, and I have mastered various tools and skills, such as FastAPI, AWS, and NoSQL. I am always eager to learn new things and challenge myself, and I aspire to become a leader and an innovator in the field of AI and ML. I believe that AI and ML can help humans to go beyond limits and create positive impacts for society.

Experience

1. Machine Learning Engineer

   Evidium

   Aug 2024 — Present · San Francisco, CA, US

   Built an end-to-end pipeline to map unstructured clinical text to structured medical codes (e.g, ICD/SNOMED) on real-world hospital data; designed an evaluation harness that enables faster iteration speed2. Led data curation using a Self-Taught Reasoner (STaR) approach and trained models via SFT and preference alignment (ORPO, DPO, KTO, APO), improving recall-precision tradeoff. Developed the RLVR training infrastructure (GRPO) for query improvement, boosting retrieval quality and outperforming Google MedNLP and Amazon Comprehend Medical in internal evaluations.4. Took full ownership of the end-to-end product, from idea to deployment, and built clinician-in-the-loop labeling infrastructure with clear guidelines, workflows, and quality checks, making it easier for our clinicians to collaborate in the development process

Education

• National Institute of Technology Rourkela

  Bachelor of Technology - BTech, Major in Mechanical engineering and Minor in Electronics and communication engineering

  2017 — 2021

• Indiana University Bloomington

  Master's degree, Computational Science