All Projects

Built a multimodal VQA system combining ResNet-152 visual features with GloVe+LSTM text encoding. Implemented layer-wise residual co-attention mechanism achieving ~60% accuracy on VQAv2 benchmark.

Developed a 3-module pipeline for generating contextual follow-up questions with iterative correctness checks. Fine-tuned on 26k immigration QA pairs, improving ROUGE scores over baseline.

Analyzed the impact of vague language instructions on Vision-Language Navigation (VLN) agents. Studied how ambiguous spatial and directional terms affect navigation performance and agent interpretability.

Research on syntactic feature encoding in Multilingual BERT for code-switched language. Layer-wise probing study analyzing how mBERT handles mixed-language text patterns.

Implemented multiple ML classifiers (SVM, Random Forest, XGBoost) for mental health condition prediction from survey data. Explored feature importance and model interpretability.

Built a boolean information retrieval system from scratch in Python. Supports AND, OR, NOT operators with inverted index construction and query parsing.

Android application for personal task and inventory management. Real-time sync with Firebase, intuitive UI for tracking items and deadlines.

Implemented core ML algorithms (Linear/Logistic Regression, Feature Selection, Gradient Descent) without using ML libraries. Focus on understanding the math behind the models.

Front-end compiler for a custom mini-language called DogeLang. Includes lexer, parser, and AST generation for basic constructs and expressions.

Fintech mobile app built for Zeta Hacks 2021. Flutter-based expense tracking and budgeting with clean Material Design UI.

Mobile-based agricultural information system designed for Indian farmers. Features crop suggestions based on geographic location and demand, fertilizer/pesticide recommendations, and timely farming assessments with an intuitive native-language UI.