Shruti Sivakumar

This project explores the inverse design of materials using a Conditional Variational Autoencoder (CVAE) framework trained on a curated dataset of Refractory High-Entropy Alloys (RHEAs). By leveraging deep generative modeling, we aim to identify novel alloy compositions with desired mechanical properties, such as high yield strength.

No description provided.

Multimodal deep learning pipeline for hateful meme detection. Combines image and text features through fusion layers for binary classification on the Facebook Hateful Memes dataset.

Real-time Network Intrusion Detection System built with Apache Spark, Kafka, and deep learning (CNN/LSTM) on the CSE-CIC-IDS2018 dataset. Extends state-of-the-art research with a live streaming detection pipeline — all running on a single machine via Docker.

Designed and compared linear, nonlinear, and deep learning time-series models for highly volatile financial data; conducted stationarity testing, residual diagnostics, and volatility analysis to guide model selection.

Benchmarking 6 MSA tools (Clustal Omega, MUSCLE, MAGUS, M-Coffee, MSA Probs, MSA Transformer) across BAliBASE, OXBench & PREFAB datasets.

Myntra MixNMatch Studio and Style Showdown by Team InnovateHers 🎨👗 [Winners of Myntra WeForShe Hackerramp 2024]

Comparative analysis of LSTM, XGBoost, Random Forest, SVR, and SARIMAX for wind power prediction using real-world turbine data. Covers preprocessing, time series modeling, and performance benchmarking.

A voice- and text-based GenAI assistant that helps government officers query public welfare scheme data using natural language. Built with Azure OpenAI, Speech Services, and SQL for real-time insights into MGNREGA, PMAY, Ujjwala, and other key schemes.

Smart text summarization with BART & Pegasus AI models. Supports text, URLs, and files. Built with FastAPI + Next.js 14. Features: dual model comparison, caching, auth, history tracking.

FlowScript is a learning-oriented language and toolkit for translating human-friendly logic into executable behavior and real programming languages. It includes an interpreter, code translators (Python, Java, JavaScript, C++), a step-by-step execution engine, a flowchart generator, and a browser-based playground.

Language Identification using Markov Chains, Hidden Markov Models, and Neural HMMs. Explores probabilistic and deep learning methods for sequence modeling, with ablation studies on context, hidden states, and optimization.

No description provided.

Predict AAPL stock prices using Lasso Regression solved via ADMM optimization. Sparse modeling + custom Python implementation.

Predict 3D protein structures using AlphaFold2 via ColabFold, with visualizations of per-residue confidence and alignment error.

An analysis of social media networks using graph theory and machine learning. Investigates Facebook verified page interactions via centrality metrics, community detection, and page characterisation using structural embeddings.

Missing data imputation techniques implemented from scratch in C — includes Linear Regression, KNN, Mean/Median, Listwise Deletion, and Hot Deck methods, tested on real-world datasets. First university project focused on practical data preprocessing.

Unsupervised anomaly detection for credit card fraud using Gaussian Mixture Models (GMMs). Models legitimate behavior via probability density estimation and flags low-likelihood transactions as potential frauds.

Built two NLP solutions for automating business memo parsing and bid title keyword classification. Includes GPT-4o fine-tuning, DistilRoBERTa multi-label tagging, and a Streamlit app for internal deployment.