Sankshepika: Legal Document Summarizer

(CS725: Foundations of ML | Instructor: Prof. Sunita Sarawagi) (Autumn '23)

Improved the pretrained LegalLED with NLTK for simplifying jargons and Indianizing legal terms. Finetuned LegalLED with Indian dataset while extracting important sentences using the TF-IDF method. Combined the summaries with OpenAI to overcome token restrictions, achieving a ROUGE score of 0.45.

Generation of synthetic tabular data to enhance classifiers for highly unbalanced data

(Jan '24 – Present)
(M.Tech Project | Guide: Prof. Sunita Sarawagi)

Increased the classification accuracy by 2.5% across GAN and diffusion models for imbalanced data. Developed ORD, fine-graining the binary target with more information on overlap in majority class. Empirically proved that ORD enhances synthesizer, but removing overlap improves classifier performance. Studied privacy preservation techniques for the generated synthetic data against common attacks.

Publication: Accepted at A* conference AAAI'25, titled "Synthetic Tabular Data Generation for Imbalanced Classification: The Surprising Effectiveness of an Overlap Class".

Exordium

This is an official Mobile Application for our college library GRD Memorial Library. It is designed for PSG techians and outsiders with an aim to describe the various resources and opportunities available at the library. This has the largest user base compared to other projects.

View on PlayStore
Platform: Flutter SDK
Programming Language: Dart
IDE: Visual Studio Code, Android Studio

Analyse IPL Auction Hashtags

This is a Social and Economic Network Analysis project. Tweets related to IPL Auction 2021 were scrapped from Twitter using TWINT tool. Hashtag counts and mention counts were analysed and visualised as Word cloud and barchart. Community detection algorithms like Louvain were applied to see the grouping of hashtags.

Source Code
Tools: Networkx, Twint, Wordcloud
Programming Language: Python
IDE: Google Colaboratory, Gephi

Weed Detection in Sugarbeet Crops

This is a Machine Learning and Computer Vision Project for locating where exactly the weeds are present when given a picture of sugarbeet field. It has good applications in precision agriculture field.

Source Code
Platform: Tensorflow using UNET Model (CNN)
Programming Language: Python
IDE: Google Colaboratory

Illumine

Book reading list and review system. It is just the app for ardent book lovers. You can keep track of the current reads, wishlists and completed books. We can also view the books and tag the Quotes.

Source Code
Tools: PHP, MySQL, Bootstrap
Programming Language: PHP, HTML, CSS, JS
IDE: VS Code (hosted through XAMPP)

Grado

This is a CGPA calculator app. It is useful for Choice based credit systems in colleges. It helps find both SGPA per semester and cumulative CGPA.

Source Code
Platform: Andoid Studio
Programming Language: Java