Machine Learning, Optimization
LLM judges are stochastic — the same prompt-response pair scores differently across queries, so uniform sampling wastes budget on low-variance pairs. We model each pair as a bandit arm, estimate variance online, and…
Theoretical ML, Math
In combinatorial bandits, a learner picks a subset of arms each round and observes a reward aggregated by an unknown function. Prior work required strong structural assumptions (e.g., linearity) on the aggregator,…
Reinforcement Learning, Deep Learning
Created an experiment to modify reward structure of reinforcement learning algorithms to enhance the learning capabilities in environments with sparse rewards. And experimentally showed that the algorithm performs…
Reinforcement Learning, Deep Learning, Fun!
Wanted to test whether an algorithm such as DQN which was able to play atari games. Could also detect when the game being played had bugs.
Computer Vision, Machine Learning
In working to detect products placed in the cart Detecting when products are put into a cart. We wanted to use all the information that we were receiving from the camera.
Computer Vision, Deep Learning
Implemented a research paper’s code that de blurred video content. However the code had an issue where the output videos occasionally had artifacts in areas that were relatively still and not moving.
Computer Vision, Deep Learning, Data Science
Enabled creation of really cool datasets of objects, using videos of objects being manipulated by hand.
Computer Vision, Deep Learning
Gathered the images of hands with the products.
Computer Vision, Deep Learning
Created a solution to capture the logo of a company from images of the products.
Computer Vision, Deep Learning
The objective was to group together similar objects based on visual appearance. The main
Computer Vision
A common problem in machine learning is using a variable number of inputs to get a fixed output.
Data Generation, Computer Vision, Deep Learning
The goal was to digitize nutrition tables from images. Manually labeling data was too expensive and time consuming.
Computer Vision
The task was to create a system that could overlay any design on top of a cloth that a model was
Data Science, Computer Vision, Fun!
This is a deep learning model that analyses photos to determine the emotional response of a human seeing the photo
Natural Language Processing, Deep Learning, Generative AI
Used ChatGPT and prompt engineering to summarize customer support conversations.
Natural Language Processing
This program searches different search engines (Google and Bing) and uses different browsers (Chrome, Firefox, Internet Explorer). Searches for phrases and counts the number of “.com” URLs that show up before the…
Natural Language Processing, Data Science, Machine Learning
From the images captured I created a system that uses it well.
Data Science, Natural Language Processing
Wanted a way to search across Dell support system. To guide customers directly to the relevant support articles. The problem was that the sentence search wasn’t context aware and keyword based.
Machine Learning, Data Science, Natural Language Processing
Wanted to create a grocery app that contained all the groceries from all the major online grocery vendors. Allowing users to choose the groceries from each store individually.
Data Generation, Time Series Forecasting, Machine Learning, Data Science, Fun!
Connected a vibration sensor to a lathe machine to predict the level of wear using the amplitude of vibrations.
Machine Learning, Data Science
An online sales company had difficulties with their returns. As their returns were cutting heavily into their profits.
Time Series Forecasting, Data Science
Created a tool using holt winter’s algo and other time series forecasting tools to create a tool that generates predictions of the amount of footfalls in Boston Children’s museum.
Automation, Machine Learning, Data Science
Wanted to create a tool that allowed non-technical users to be able to gather insights from the data from their company and effectively automate the job of a data analyst.
Automation
Created a program to manipulate xlsx files.
Automation
The requirement was to find if the page distorts, if zoomed to 200%.
Automation
The requirement was to save the NVDA spoken history. This was not available since the Software didn’t have that option and there wasn’t any add-on to save history.
Fun!
Created a sudoku solver. It could take any sudoku puzzle and solve it as long as there was at least one tile where only one number was possible.