Saurav Kumar

I'm an undergrad at UIUC studying CS, Stats, and Physics. I'm from the bay area.

I'm currently researching LLM synergies and adding complexity to the knowledge side of LLMs. I'm exploring the psychology/neuroscience of consciousness and creativity and replicating it with chain-of-thought strategies. I'm also obsessed with astronomy, and it's greatly influenced my thoughts on Simulation Theory and AGI.

I tackle problems with curiosity and ambition. I strongly believe my childlike wonder on ideas I explore makes me 10x more effective and fuels my creativity. In the past, I've spent time building Federated ML systems at Stanford Medicine and AMD (through UIUC disruption lab) – a distributed ML strategy for edge-computing. I am currently learning about Robotics RL, neuroscience, and biological machines.

I run an AI research program called BLAST AI. We train high school students on ML and put them through a research project. This summer, we took in 105 students and ran 15 different research projects ranging from Molecular LLMs to using LLM skill libraries for RL. Giving real trust to students with incredibly diverse backgrounds has been game-changing for our research. We've taught highschool CERN researchers and Olympiad winners from places like Ukraine and Nigeria.

In my free time, I love playing basketball, finding lakes, and becoming an influencer on Twitter.

If you have any ideas or want to talk, please reach out at epicsaurav [at] gmail.com

Projects and Research

Macrocosm: a group embedding the internet. Worked on a local fine-tunable chatbot with quantized versions of LLAMA. The group received millions of views on Twitter.

Building a molecular LLM and semantic search database for molecules (researching with BLAST)

SMLC: A course on competative ML (27k students on Udemy)

Where's Bora: hacked FindMy to make a website tracking my friend's location 24/7 (with consent).

An NLP strategy to classify medical specialties from normalized patient phone transcripts – published in IEEE.

CV for airplane contrails & atmospheric chemical waste – published alongside NASA in AIAA

Announcr: product placement ad marketplace for TikTok – received a VC grant and influencers with 11 million followers.

Lapis News: a positive news app and ML algorithm published in IEEE (charted on the App Store!)

Planet 9: a book on planetary science and a theoretical 9th planet

A ton of failed financial ML models including meta-labeling pairs mean reversion and Twitter sentiment analysis for crypto. I'll find a strategy that works one day!

Writing