1. Duality of Generative & Discriminative Recommendation (DRAGON) under Prof. Julian McAuley, UCSD (Oct 23 - Present)

   1. Proposed a novel approach for the task of sequential recommendation which include the benefits of both the generative and discriminative paradigm to achieve SoTA results (increment of ~5% in nDGC@5 in comparison to the current best model) on Amazon’s recommendation dataset of Instruments, Games, Baby and Office products. 

   2. Paper on the above submitted to NuerIPS 2024.

2. MathNet under Prof. Neil Heffernan, Worcester Polytechnic Institute  (Aug 22 - Jul 23)

Working on auto-grading of open-ended math problems from student-uploaded image answers using techniques such as document analysis. 

One of 2 teams within an application pool of 10 teams to receive a $100K grant from the Jaffe foundation, founded by Laurence Holt and Doug Jaffe, to further pursue research in this area.  

Published two papers - 

2. 2. 1. Auto-scoring Student Responses with Images in Mathematics - EDM 2023

      2. Investigating Patterns of Tone and Sentiment in Teacher Written Feedback Messages - AIED 2023  

3. Logistic Innovation Lab under Prof. N.S Narayanaswamy (August 2021 - July 2022)

Lead a team that built products to extract vehicle and driver information from paperwork, detect unsafe driving, dynamically redirect vehicles based on factors like congestion, weather, and estimate ETA using Computer Vision and Machine Learning.

4. Creativity in AI under Prof. Krishnan Balasubramanian (June 2021 - Dec' 2021)

Built a novel model using a combination of two transformer-based models to encode the style of a professor and generated lecture transcripts on theoretical topics with the style of the professor. Secured the highest grade ‘S’ for my research contributions to the project. 

5. Players Prediction in the IPL Auction under Prof. Rupesh Nasre (January 2020 - April 2020)

             Worked on an algorithm with minimal runtime to deterministically predict the best combination of players given budget constraint from a large existing  dataset.

    Solved the problem and reduced the runtime by 100x using a combination of dynamic programming and Dijkstra’s algorithm.

Link: https://github.com/AbhishekSanthanam/IPL_Auction_Best_Combination_Predictor_DBMS