Engineered an AI-powered accessibility tool bridging speech, text, and sign language, significantly improving real-time communication and accessibility for individuals with speech or hearing disabilities. Designed and implemented speech-to-text, text-to-speech, and text-to-sign language translation using advanced deep learning techniques, NLP models, and computer vision algorithms to enhance accuracy and usability. Developed a browser-based extension for seamless accessibility and real-time translation.

Optimized real-time emergency response and resource dispatch by integrating Folium for live geospatial tracking, allowing responders to be dynamically assigned based on proximity and availability. Ensured high availability and concurrent request handling by developing an asynchronous FastAPI backend with PostgreSQL, enabling seamless incident logging, updates, and retrieval. Enhanced decision-making speed by streamlining incident-classification and using an OpenAI powered chatbot. Improved system security and scalability with role-based access control (RBAC), API rate limiting, and multi-factor authentication, ensuring controlled data access while maintaining performance.

Engineered a Deep Convolutional Generative Adversarial Network (DCGAN) for creating high-quality anime character faces. Achieved exceptional results with a generator loss of 5.6672 and discriminator loss of 0.0950 over 25 epochs. Implemented advanced activation functions and architectural optimizations, resulting in 94.34% accuracy for real image detection and maintaining a low 2.64% false positive rate for generated images.

Developed a sophisticated CNN-based emotion recognition system using the FER2013 dataset. Implemented comprehensive data preprocessing pipeline including image normalization and grayscale conversion. Achieved 63% accuracy on the test set with detailed performance analysis using multiple metrics. Conducted in-depth analysis of model behavior and identified potential optimization strategies through systematic evaluation of misclassified cases.

Conducted comprehensive analysis of Air Quality Index (AQI) using data from the Central Pollution Control Board. Leveraged ARIMA modeling to forecast pollutant levels and their impact on air quality. Generated actionable insights and recommendations that could lead to a 5% improvement in AQI, based on correlation analysis and thorough literature review of environmental impact factors.

Designed and implemented a deep learning model using Keras for predicting customer churn in the telecom sector. Executed comprehensive data cleaning and transformation processes to ensure high-quality model inputs. Achieved 78% prediction accuracy on unseen data, with detailed performance evaluation using multiple metrics to support the development of targeted retention strategies.

Analyzed terminologies and trends in genetic research using Natural Language Processing (NLP) techniques. Identified key genetic traits by processing scientific abstracts, applying tokenization, stop-word removal, and phrase extraction and generated word clouds, visualizing the most relevant terms in the dataset. Developed a document classifier with BERT, achieving 75% accuracy in predicting research paper relevance for QTL curation and analyzed terminology relationships by training a Word2Vec model.