• Trained machine learning and Transformer models using TensorFlow to predict bidding prices in a supply chain industry, achieving a 30% reduction in prediction error and potentially saving 5M+ dollars annually for the client in shipment costs• Conceptualized and led the development of an end-to-end LLM-powered virtual teaching assistant for an Ivy League business school, achieving daily engagement from 1000+ students. Collaborated with client’s cross-functional teams to ensure solutions adhere to organizational standards of data governance, role-based user access, and establishing guardrails• Deployed scalable Docker containers and API endpoints on Microsoft Azure to establish a robust data engineering framework. Leveraged Azure Tables to build a real-time Change Data Capture (CDC) pipeline consolidating data from 10+ sources• Text to SQL Translation System: Fine-tuned Llama models to translate natural language queries into SQL commands. Achieved 75% accuracy in query disambiguation and 90% in context understanding, with an 8-second average response time. Enabled user-friendly interaction with structured databases for non-technical users, reducing data access time by 80%• Led and mentored a team of University of Chicago students in their Master’s thesis on LLM-based Recommender Systems

• Enhanced speech recognition for young children by 30% by leveraging LoRA and quantization to fine-tune Whisper models. Achieved a 75% reduction in model size to optimize for edge devices with limited hardware while maintaining high accuracy• Conducted comprehensive data analysis on over 1,000 audio transcripts, identifying key patterns in more than 10,000 spoken words using topic modeling and NLP techniques. Mapped these insights to specific stages of language development, enhancing understanding of early childhood linguistic progression and informing targeted intervention strategies

• Implemented Graph Convolutional Neural Networks from scratch using tensorflow to predict important features of drug molecules, resulting in a potential 30% reduction in experimental cost and time for the client• Collaborated with a leading global pharmaceutical company to research and identify potential drug candidates using machine learning; Research to be published in a top journal soon• Created a ReactJS-based web application with neo4j as the backend, to enable the efficient and real-time search of 100M+ text documents• Managed and mentored a 5-member team by writing tutorial blog posts, designing quizzes, and collating study materials for data science roles, shortening the onboarding time from 6 to 2 weeks

• Developed multiple data science POCs to evaluate potential AI applications in various domains leading to successful client meetings and new business opportunities• Improved client customer service by automating the classification of incoming customer emails using NLP and self-attention neural networks, reducing the ticket generation and response time from 2 days to a few minutes• Built a forecasting model using a custom multi-output recurrent neural network that accurately captures temporal dependencies in multivariate time series data. Achieved a 50% reduction in error compared to traditional RNNs• Built an Airflow-based framework to visualize and monitor the ETL pipeline with 30+ docker containers, resulting in a streamlined process and optimized pipeline management

• Added a word-prompt feature for users while typing, resulting in better user experience and productivity• Trained initial models for a QA chatbot that led to a 10% reduction in the number of customer calls down the line