Aniket Malpure

Pioneered a Speech Emotion Recognition and Voice Analysis System leveraging the state-of-the-art wav2vec2 model to accurately map emotions and empathy in classroom settings, significantly enhancing faculty productivity.

Engineered and deployed pre-trained Audeering and SpeechBrain models to extract critical voice features from over 20 faculty member audio responses, establishing a robust dataset and boosting emotional analysis accuracy by 5%.

Developed a Fuzzy Name Matching Algorithm utilizing 20 string similarity features and Deep LSTM Siamese networks, elevating customer identification accuracy by 11% (from 83% to 94%) across critical banking applications.

Architected a robust stacked Ensemble Model, integrating AutoML-optimized base classifiers with character embedding networks and Logistic Regression Meta-Models via Stratified 10-fold Cross-Validation, significantly reducing manual verification efforts in loan applications.

Modernized and optimized PostgreSQL database architecture for Product on Demand (EMI POD, Document POD, PartnerLending POD) platforms, resulting in a 30% improvement in response times and a 15% increase in operational efficiency.

Automated critical data workflows by designing and implementing ETL/ELT pipelines using Azure Data Factory for PostgreSQL, streamlining PowerBI reporting, reducing manual effort by 25%, and ensuring timely insight delivery.

Engineered and deployed CI/CD pipelines with Azure DevOps for PostgreSQL databases, integrating Azure Key Vault, Azure PowerShell, and Azure Repos to facilitate secure and efficient deployments.

Recognized with a Kudos award for excellence in maintaining comprehensive Confluence documentation, fostering enhanced team collaboration and project success.

Developed a CNN-based Plant Disease Classifier, achieving high performance metrics: 95.62% accuracy, 94.38% precision, and 93.60% recall for robust disease detection.

Utilized transfer learning techniques with VGG19 and ResNet architectures to benchmark performance and optimize model accuracy for plant disease classification.

Performed advanced Sentiment Analysis using NLTK, VADER, and ROBERTa, identifying key behavioral patterns that led to a 15% increase in user engagement.

Engineered a sophisticated Recommendation System, integrating content-based and collaborative filtering with Graph Neural Networks (GNNs) for user-item relationship analysis, driving a 30% improvement in user engagement and content accuracy.

Optimized real-time SQL queries, significantly reducing execution time by 120ms and enhancing overall database performance to support improved decision-making.