Engineer at UrbanCompany. Just exploring tech...

Sarthak Madhamshettiwar  • Peerlist

Implemented a model that aims to predict whether the input tweet is about a real disaster or not Performed EDA, and pre-processing techniques to clean the raw data. Used GloVe Embedding to convert textual data into numerical format. Used Classification algorithms and LSTM based Neural Network for binary classification Employed BERT, for generating sentence embedding of input, modelled Multi-layered Dense Neural Network for final classification. Achieved validation accuracy of 81.8% and public F1 Score of 82.80%

Tweet Disaster Prediction

Developed a Quote Generation Model, that generates quote using the name of anime character and a few initial words. The generated quotes are unique, and represent the point-of-view of the character Fine-tuned Google Gemma-2B with training loss of 59.9%. Optimized the process of finetuning using QLoRa, PEFT and SFT trainer from HuggingFace for efficient finetuning Applied preprocessing techniques using RegEx and visualized tweets using WordCloud for better analysis of output

AnimeGPT

Developed a web-based robust Blogging app, where user can add new blogs, read existing blogs, remove/update their previous blogs (CRUD operations), and comment on any blog Leveraged JWT(JSON Web Tokens) to implement session management and user authorization and authentication to secure user accounts, ensuring each user’s credentials remains private and protected Reduced time taken to render frequent blogs by 41% by integrating Redis for caching frequently visited blogs Used BullMQ with SendGrid to notify users whenever their blogs get published on their respective emails Integrated the application with MongoDB database to store and retrieve blogs and comments and user details