Data Scientist

Education

• M.Sc. in Economics, BITS Pilani (May 2014)
• B.E. in Chemical Engineering, BITS Pilani (May 2014)

Publications

HALT: Hate Speech Alleviation using LLMs and Transformers

Technical Skills - Python, SQL, Tableau, AWS, GCP, Docker

Key Projects

Natural Language Processing (NLP) - LLM-Based

• Uber Data Platform Insights: Analyzed over 5000 user conversations to uncover 5 major pain points, driving actionable product enhancements that improved user satisfaction. Insights were integrated into the platform for long-term improvements.
• Uber Freight Data Assistant: Built an automated response system addressing ~60% of frequently asked queries by integrating data from multiple internal tools, reducing manual effort and enhancing operational efficiency.
• LLM Chat Agent: Designed a chat tool leveraging OpenAI, LlamaIndex, and Streamlit, integrating open-source tools for search, weather updates, and code execution.
• Data Whisperer: Engineered an LLM-based chatbot for dynamic data interaction, showcased at a hackathon, demonstrating innovative use of language models to drive data-driven conversations.

Classical NLP

• Chat Abuse Detection: Developed a custom NLP model to classify abusive content in Indic languages transliterated in Roman English Script, achieving an F1-Score of 97%+.
• Automated Bug Scoping: Automated manual scoping using ML techniques, achieving a 10x reduction in scoping times for applicable bugs.

Computer Vision (CV)

• Fine-Tuning Stable Diffusion: Implemented ControlNet on Stable Diffusion using a custom Pokémon dataset. Managed the entire project lifecycle, including converting images to canny edges, generating descriptions, and fine-tuning the model. Checkpoints were created, and outputs were monitored at various steps for optimization.
  Project Link
• Custom Image Detection - YOLO: Built a custom image detection model to identify landmarks in Hyderabad, India. Handled end-to-end tasks including data annotation, model training, and evaluation.
  Project Link

Intelligent Systems

• Mastermind Anomaly Analysis: Designed and implemented an automated system to analyze anomalies in Mastermind, Uber’s tool for detecting fraudulent user behavior. The system identified root causes of alerts by analyzing feature drift, user behavior changes, or genuine suspicious activity. Automated analysis streamlined workflows and provided actionable insights for stakeholders.

Data Analytics

• Safety Experimentation: Spearheaded the analysis and readout for Uber’s 4W Safety Education program in Latin America. Designed and conducted experiments to promote safe driving behaviors among earners, delivering insights to refine the program and boost engagement.
• Scoping Accuracy Improvement: Led efforts to improve bug scoping accuracy for the Rider Line of Business. Increased accuracy from less than 30% to approximately 80% through targeted analysis and process improvements, enabling efficient prioritization of impactful bugs.
• Uber Data Platform (UDP) Analytics: Partnered with the Uber Data Platform team to design a scalable analytics solution. Focused on process optimization, product improvements, and governance, enabling teams to identify inefficiencies and implement targeted enhancements.