xLSTMTime – Time-Series Forecasting
Implemented the xLSTMTime architecture combining sLSTM and mLSTM with exponential gating and enhanced memory for long-term multivariate forecasting. Benchmarked on electricity demand and traffic datasets with 5–10% lower MSE/MAE than standard LSTM baselines and outperforming transformers on select horizons.
PyTorchsktimeTime SeriesLightningDataModuleAttention
Tversky-Layer Implementation
Implementation based on 'Tversky Neural Networks: Psychologically Plausible Deep Learning with Differentiable Tversky Similarity'. Provides a Tversky-based layer for controllable similarity learning in deep models.
Deep LearningSimilarityPyTorch
Anime Face Generation (DCGAN)
DCGAN trained to synthesize realistic anime faces. Generator/discriminator trained adversarially; results show crisp features and stable training with tuned learning rates and regularization.
GANDCGANComputer VisionPyTorch
Image Caption Generator
End‑to‑end pipeline on Flickr8k using VGG16 for feature extraction and an LSTM decoder with attention to produce context‑aware captions. Achieved BLEU‑1/2/3/4: 0.4851 / 0.3028 / 0.2084 / 0.1013.
TensorFlowKerasVGG16LSTMAttentionNLP
Neural Machine Translation
Seq2Seq encoder‑decoder with attention for translating between languages. Trains on parallel corpora and demonstrates fluent target‑language outputs from fixed‑length latent representations.
NLPSeq2SeqAttentionEncoder‑Decoder
Answer‑IT
Tool for quick, concise product answers by streamlining retrieval of product details and returning instant summaries.
AIRetrievalSummarization
OptiFuse
A next‑gen approach for minimizing server costs in serverless platforms using fusion functions. Built with my friends Arijit Konar and Vaivaswat Dubey.
ServerlessCost OptimizationFunctionsPerformance
