I am an engineer and researcher from New Delhi, India
Think of this as a catalogue of everything I’ve been up to
I work full-time as a Consultant Specialist (ML / Backend) at HSBC and part-time as a Research Assistant at IIIT Delhi
Check out some of my work below
Festigo | Cross Platform Social Mobile App
Festigo is an app that enables people to find dates and friends based on shared interest in live events. A curated feed of live events is shown, where you can view details about each event and choose the ones you like. Then you can match with other users interested in similar events in the “Date” and “Buddy” feeds and chat with those you match with. Festigo has been developed using Flutter with a Firebase backend. Features like user matching, event filtering, chat, etc. are all fully functional.
Mobile Development, UI/UX, Asynchronous Programming, Entrepreneurship, Generative AI
Generating Constructions for the Guillotine Cutting Problem in Combinatorial Geometry using Reinforcement Learning
Some combinatorial problem spaces are so complex that it is impossible for a human to see patterns in them, let alone adopt a logical strategy to build constructions that satisfy quantifiable conditions. The Guillotine Cutting Problem in combinatorial geometry is one such problem. In my research assistantship with Prof. Rajiv Raman at IIIT Delhi, we are developing a reinforcement learning agent that aims to generate constructions of rectangles for which an upper limit for killed rectangles in an optimal cutting sequence is breached. This is an extension of the method proposed by AZ Wagner in his paper Constructions in combinatorics via neural networks (2021).
Research, Deep RL, Dynamic Programming, Theoretical Mathematics, Recent
Upaj Lekha | Multi lingual Mobile App for AI-Automated Agricultural Record Keeping
Upaj Lekha translates to “Crop Journal”. This explains its core function, which is to replace the pen-and-paper record keeping process prevalent amongst Indian farmers. Upaj Lekha’s smart reader feature lets farmers take pictures with their phone for automatic record taking. Text is extracted from the picture using OCR and relevant fields are captured using NLP techniques. This makes record keeping accessible to illterate farmers. The app contains user authentication via mobile number, persistence on firebase, farming advisory, waste management and other features. Developed using Flutter.
AI, OCR, Natural Language Processing, Mobile App Development, UI/UX, Social Work, Entrepreneurship, Recent
Publication | Identifying Offensive Content in Social Media Posts using Fine-Tuned DistilBERT
The identification of offensive content in social media is a widely studied field. At the time of this publication, the most performant methods used for this task were resource-intensive and trained on relatively small datasets. We used both traditional machine learning methods and state of the art transformer models like BERT to set a baseline for our experiments. Following this, we propose the use of Distilled Bert fine-tuned using both OLID (from SemEval 2019) and an additional hate speech and offensive language dataset. We were able to successfully demonstrate competitive performance using a lighter model, with a more general training set.