Working on M365 Copilot's AI reasoning and agent capabilities. Key contributions:

• Built the Analyst Agent — prompt-engineered Python Code Interpreter (CI) integration that improved CI trigger rate by ~100% and user satisfaction by ~150%.
• Improved numerical citation rendering for CI output, yielding a 40% improvement in P99/P95 ChatLTR.
• Drove GPT-5 transition; prompt structure optimizations reduced ChatFTR by 30% across performance tiers.
• Developed synthetic data generation pipelines for fine-tuning reasoning models across code generation, data analysis, web search, and multi-site crawling use cases.
• Built synthetic tenant generation systems producing realistic enterprise environments (emails, meetings, calendar events) for agent training and evaluation.
• Implemented agentic graders for the Outlook Agent to detect hallucinations and measure inbox coverage.

LLMsPrompt EngineeringAgent SystemsSynthetic Data

Minor in Data Science and Machine Learning  ·  CGPA: 9.04 / 10

Worked on the Smart Subs team, recommending substitute items when products go out of stock. Trained XGBoost and Neural Network models to rerank substitutes based on user preferences from historical data. Improved quantity ratio logic to account for item pricing, achieving a 10% reduction in average price of substitutes recommended to customers.
MLRecommender SystemsXGBoost