Work
LDC Mixes
Looking for just the right vibe for your upcoming function? Look no further! Whether your event is just starting, at its peak, or somewhere in between, we've got the vibe for you. Presented by Jaxon Clay and LDC.
Walking In » | Turn Up » | Absolute Bangers »
app.saarim.me
I build a variety of software tools available at app.saarim.me.
Some notable things I've built include a ChatGPT-like chatbot that lets you chat with anyone's Twitter persona, a prompt improvement tool, and a bill splitter
that parses receipts.
MedBytes
MedBytes is a quick and efficient way for doctors to stay on the cutting-edge by providing the highest impact research and news for their respective specialty. Delivered via a fresh, personalized, and easy-to-navigate user interface, MedBytes’ mission is to offer physicians an all-in-one platform for their medical content that will enable them to better serve their patients.
Check out my talented co-founders — Jeff Diamond, Sid Kaushik, and Vedaad Shakib!
ML Research
As part of my work with BAIR, I've explored a breadth of topics relating to deep neural networks (DNNs). You can view my exploratory research (and some cool plots) in the notebooks below.
DNN Ensembles: Part 1 » | Part 2 »
Outlier Detection: Supervised Isolation Forest » | DNN Embeddings »
Unsupervised Learning: UMAP Loss »
Acta
Acta aims to build a fake news detection algorithm as a service that will allow anyone to track the propagation and veracity of any tweet. We focus on the following goals: provide meaningful analytics on the information landscape of social networks; detect disinformation by analyzing its propagation; strive to alleviate some of the repercussions of fake news.
Check out my talented team members — Vincent Liu and Jupinder Parmar!
Implementing ML Algorithms
For the Spring 2021 offering of Berkeley's CS 189, I implemented a variety of popular machine learning algorithms without using any libraries for out-of-the-box classification (e.g. sklearn). I would submit my models' predictions to Kaggle for an in-class prediction competition.
I placed in the top 3% and top 7% of my class of over 500 students for my convolutional neural network and Gaussian discriminant analysis implementations, respectively.
Algorithms: Gaussian discriminant analysis (linear and quadratic discriminant analysis), RIDGE logistic regression, decision trees, random forests, bagged trees, feed-forward fully-connected neural networks, convolutional neural networks.
Coping with NP-Completeness
For the Fall 2020 offering of Berkeley's CS 170, my team and I were tasked with solving and approximating solutions to an NP-complete problem. You can view the problem statement and project spec here. We chose to formulate this problem as an integer linear program, which we found solutions to using Gurobi.
We placed in the top 8% of our class of over 600 students.
Check out my talented team members — Vikram Shirsat and Kevin Xu!
Crypto Sentiment Trader
Developed a trading bot that independently identifies and places favorable trades on Binance by parsing real-time financial metrics and evaluating social media sentiment on high-volume cryptocurrencies such as BTC, ETH, and LTC.