I solve hard problems.
I want to work in a place doing exactly that, solving the world's most pressing issues alongside smart, passionate people who feel that math and machine learning can make the world a better place.
Here are some of the toughest challenges I've worked on:
ML Research: I recently published original research on tabular ML – A Case for k-Means Binning for Histogram Gradient-Boosted Decision Trees – in Transactions on Machine Learning Research (TMLR), becoming among the journal’s only solo undergraduate authors. This paper blends theoretical backing with extensive empirical research, and was recently chosen to present at ICLR (2026) as part of the top 10% of papers published in 2025. I'll also be presenting it at the AI for Tabular Data workshop at EurIPS (2025) in Copenhagen.
Semantic Embeddings: As the Principal AI Researcher at AkashX, I developed a novel method for integrating semantic retrieval directly into databases to maximize speed while ensuring epsilon-accurate predictions even with low-parameter embedding models. An advancement on the LOTUS paper published in mid-2024 from both a mathematical and engineering perspective, our architecture is currently awaiting patent approval.
LLM Development: In my current Brown-funded research, I’m training a series of recurrent 1B-param SLMs to analyze how token- and sequence-based recurrence affect reasoning capabilities. This has required me to learn the entire stack of LLM creation, from multi-gpu distributed training to random matrix theory to dynamic dynamo compilation. While my repository is currently private, I’m happy to share more details upon request.
ML in Production: For the 2024 elections, I founded 24cast.org, the first ML-based, open-source election prediction website that combined various statistical models to predict every Congressional, Presidential, and Gubernatorial race in the nation. For election night, I built another open-source model that used conformal prediction and linear regression to forecast results for counties whose outcomes were not yet finalized. Both models required extensive production engineering to ensure robustness against a constantly-shifting data and political environment. My team, which began as just me, soon grew to more than 30 people, and our event saw more than 2000 attendees including the Mayor of Providence.
I want to work in a place doing exactly that, solving the world's most pressing issues alongside smart, passionate people who feel that math and machine learning can make the world a better place.
Here are some of the toughest challenges I've worked on:
ML Research: I recently published original research on tabular ML – A Case for k-Means Binning for Histogram Gradient-Boosted Decision Trees – in Transactions on Machine Learning Research (TMLR), becoming among the journal’s only solo undergraduate authors. This paper blends theoretical backing with extensive empirical research, and was recently chosen to present at ICLR (2026) as part of the top 10% of papers published in 2025. I'll also be presenting it at the AI for Tabular Data workshop at EurIPS (2025) in Copenhagen.
Semantic Embeddings: As the Principal AI Researcher at AkashX, I developed a novel method for integrating semantic retrieval directly into databases to maximize speed while ensuring epsilon-accurate predictions even with low-parameter embedding models. An advancement on the LOTUS paper published in mid-2024 from both a mathematical and engineering perspective, our architecture is currently awaiting patent approval.
LLM Development: In my current Brown-funded research, I’m training a series of recurrent 1B-param SLMs to analyze how token- and sequence-based recurrence affect reasoning capabilities. This has required me to learn the entire stack of LLM creation, from multi-gpu distributed training to random matrix theory to dynamic dynamo compilation. While my repository is currently private, I’m happy to share more details upon request.
ML in Production: For the 2024 elections, I founded 24cast.org, the first ML-based, open-source election prediction website that combined various statistical models to predict every Congressional, Presidential, and Gubernatorial race in the nation. For election night, I built another open-source model that used conformal prediction and linear regression to forecast results for counties whose outcomes were not yet finalized. Both models required extensive production engineering to ensure robustness against a constantly-shifting data and political environment. My team, which began as just me, soon grew to more than 30 people, and our event saw more than 2000 attendees including the Mayor of Providence.