Shanai Nair
About
Hi, i'm Shanai! I recently graduated with a BSc in Computer Science from Durham University, where I ranked second in my cohort. I have a strong passion for machine learning and artificial intelligence, and intend to enroll in a related master’s program. I am scheduled to begin an Applied Science (research) internship at Amazon in early 2025, am interviewing for additional research roles in the summer, and currently spend my time growing a "startup" that I founded.
My research interests include deep generative modeling for drug and materials discovery, and reinforcement learning on large pretrained models to enhance their capabilities and alignment. Indeed, my most recent side project involved training a 124M parameter autoregressive transformer from scratch and improving its reasoning capabilities using a separately trained process reward model.
Education
Durham University - BSc in Computer Science (2021 to 2024)
Ranked 2nd in cohort overall, and:
- 1st in Deep Learning
- 3rd in Reinforcement Learning
- 4th for dissertation (Hierarchical Transformers)
- Received an award for outstanding achievement
Haberdashers' Boys' School (2008 to 2020)
A Levels in Mathematics, Further Mathematics, Physics, and Chemistry
Extended research project prize winner
Experience
Ballo – Founder (2024 to present)
Ballo (ballo.cards) began as a side project but has demonstrated compelling financial potential. Regardless, I yearn to solve more technically challenging and meaningful problems, and primarily view the website as a means of funding my graduate education.
Amazon - Alexa AI - Software Development Engineering Intern (2023)
Interned within the Natural Language Understanding group and received a return offer.
Featured Work
This section contains some of my work from the recent past. Code is available upon request and is not public by default due to university restrictions.
Dissertation - (2023 to 2024)
Description: My final-year university project was a self-proposed research project which involved designing a novel transformer-based architecture that aimed to learn semantically meaningful hierarchical representations from its training data. I learned more from working with my supervisor on this project than anything else at university.
Mark: 85% (4th in cohort)
Feedback: Extraordinary work for an L3 project.
Link to paperDeep Learning Coursework - (2023 to 2024)
Description: The deep learning coursework involved implementing discriminative and generative models for CIFAR100 given parameter and training time constraints.
Mark: 98% (1st in cohort)
Feedback: Overall, this is a perfect submission. The report and solution are perfect, with a perfectly performing classifier and perfect realism and perfect diversity for your generative model.
Link to paperReinforcement Learning Coursework - (2024)
Description: The reinforcement learning coursework involved implementing a reinforcement learning agent to tackle the easy and hardcore BipedalWalker-v3 environments in OpenAI Gym.
Mark: 95% (3rd in cohort)
Link to paperFootprint - (2022)
Footprint was a journaling app that allowed users to upload text, audio, and video entries — before Apple offered the same within their own native app. I developed it as an independent side project during my first year of university. It has now been taken off the App Store since I do not have time to maintain it.