I like computers and the notion of computation. The idea that lot of complex systems can be described as computational processes(often of equivanent capability) and even simple computations over long period of time gives rise to immense complexity (for instance, cellular automata) fascinates me.

I am interested in building machine learning systems that can write complex computer software and use computer in a general sense. Assuming computers are the most powerful tool we have, I think machine learning systems that can write programs will allow us to explore computational universe at scale.

Most recently, I was responsible for shipping Ghostwriter at Replit and continue to develop AI models that can write code there. Before Replit, I was an undergraduate researcher at UNC NLP Lab working on code generation and code search. I'm still an undergrad at UNC and am graduating soon.

Publications

Analysis of Tree-based Input and Architecture for Code Generation
Samip Dahal, Adyasha Maharana, and Mohit Bansal
Findings of ACL 2021 [paper, code]

Scotch: A Semantic Code Search Engine for IDEs
Samip Dahal, Adyasha Maharana, and Mohit Bansal
DL4C Workshop paper at ICLR 2022 [paper, code]

Projects/Work I've done

• Ghostwriter: Replit's in-browser coding assistant.
• Scotch: An In-IDE semantic code search tool.
• Codeon: A code generation model for Python and Javascript, augmented with code search.

Some fun classes I've taken

Program Synthesis and Verification, Machine Learning, Deep Learning (Graduate Level), Advanced NLP: Structured Prediction (Graduate Level), Advanced NLP: Commonsense Reasoning (Graduate Level), Programming Language Concepts, Advanced Calculus, Introduction to Probability, Linear Algebra, Micro/Macro Economics.

Ideas I am Exploring

• In-context learning and learning from few examples
• Building machine learning models like open-source software.

Apart from computer science, I also love physics.