• Trained DeBERTa models for end-of-conversation detection with 87% accuracy.• Developed Voice Activity Detection (VAD) system to improve conversational flow.• Integrated fine-tuned models into healthcare applications, boosting patient engagement.

• Developed a chatbot using Amazon Titan Text Express v1, resolving extended token limit issues.• Tested and optimized 30-40 prompts across 50 languages, enhancing conversational outputs.• Improved PhoneCallGPT by fixing 50+ bugs, boosting call reliability and user experience.

Deployed Small Language Models (SLMs) like DistilBERT, MobileBERT, and TinyBERT on edge devices (Raspberry Pi, Jetson Nano) to enable AI in constrained environments.Optimized TinyBERT inference latency on GPU to 412.8ms, reducing memory usage by 55% and improving power efficiency.Benchmarked models, achieving 0.824 accuracy and 0.827 F1-score for DistilBERT on GPU, with a reduced inference time of 20.29s.

Organized Campus-Wide Workshops: Led the organization of workshops and meetings focused on game theory concepts, such as the Prisoner's Dilemma, and explored their real-life applications, fostering a deeper understanding among participants.Coordinated Stratathon Event: Successfully coordinated a team of 56 members to plan and execute Stratathon, the society's flagship event, which attracted over 150 participants. The event featured a series of game theory challenges, enhancing participants' strategic thinking and problem-solving skills.

During my internship at EY, I was involved in the following roles and responsibilities:Developed a highly accurate stacked LSTM model for future sales volume prediction based on previous timeseries data.Utilized a seven-layer stacked multivariate LSTM model to extract patterns from sales volume and price data.Formatted the dataset as time series data, creating 'x' (features) and 'y' (targets) for LSTM modeling.Implemented a 3-layered stacked LSTM model with mean squared loss and the Adam optimizer, achieving impressive training and validation loss results.Conducted research to enhance forecasting capabilities by engineering features from data patterns.Analyzed and interpreted results to demonstrate the model's predictive accuracy and effectiveness in sales data analysis.