August 2024 - December 2024

Domain: AI/ML, AI Observability, Synthetic Data Generation, LLMs

Business: TCS Enterprise Cognitive Business Operations

Task: Provide Proof-of-Concept to generate synthetic data generation using LLMs, to train Machine Learning models.

My Role:

• Led a team of 4 engineers, delivered Synthetic Data Generation using LLMs, recommending Claude-3.5 with 86% ML utility accuracy
• Conducted evaluations of GPT-4, Llama-2, Claude-3.5, Gemini, leveraging metrics like KL Divergence to identify best performing model
• Evaluated fidelity, privacy and utility of generated synthetic data using metrics like membership inference score, Wasserstein Distance adhering to Responsible AI
• Validated prompt engineering techniques, with Chain-of-Thought improving statistical accuracy, reasoning, and diversity in SDG
• Led Scrum calls to bridge communication between data scientists and the client, aligning on deliverables and ensuring stakeholder buy-in for the recommended solution, Claude-3.5
• Documented entire ML Lifecycle, delivered a KT session, handed over the product documentation, ensuring seamless client adoption