Amina Ashraf

Developed a comprehensive solution for analyzing customer feedback from social media by leveraging Small Language Models (SLMs) to extract key highlights and generate summaries. Implemented BERT-based classifiers to identify customer service utilization and categorize feedbacks. Successfully integrated the feedback analysis system with NoBroker's in-house CRM platform, enhancing customer insights and operational efficiency. Enhanced the embedding-based recommendation system for rental properties, achieving a 50% improvement in generating and delivering leads to users based on their activity and interactions on the platform. Built ASR system for Indic Language calls, achieving a WER of 20% by fine-tuning the Conformer model. Engineered a pipeline for identifying Indic languages from Romanized scripts, achieving ~90% F1 score. Developed a customer research module to extract actionable insights from agent-customer calls, utilizing an LLM-based chatbot built using Vertex AI. Improved the key moment identification framework in calls by leveraging LLMs and prompt engineering techniques. Worked on model optimizations and deployment, delivering on-demand APIs and real-time processing pipelines.

Developed and deployed an algorithm to identify locations with high vegetation using satellite data from ISRO. Implemented a Kernel Density Estimate based model to quantify footfall for major cities in India. Developed a C-RNN model to detect instances of negative emotions during agent-customer calls.

Developed and deployed a comprehensive web application for managing data of epilepsy patients. Designed and implemented the User Interface using HTML, JavaScript, and CSS, ensuring a user-friendly experience. Utilized MySQL for robust data storage solutions and employed Spring Boot for efficient backend development.