Navneet Sonak

About Navneet Sonak

As a senior data scientist at Drizly-Uber, I apply my machine learning expertise and data…

Experience

1. Staff Machine Learning Engineer, Llm

   First American

   May 2024 — Present

   Architected and fine-tuned the PaliGemma model using QLoRA and PyTorch Lightning, achieving higher accuracy than GPT-4.0 in data extraction task by leveraging advanced optimization techniques and resource-efficient workflows- Designed and implemented a synthetic document generation pipeline with Albumentations, OpenCV, and Pillow, simulating real-world variations like paper texture to create high-quality truth data for fine-tuning and improving model robustness- Developed and deployed DSPy to production, an automated prompt engineering framework, to dynamically add few-shot examples to system prompts, boosting extraction and decisioning accuracy for underwriting workflows- Optimized LLM prompt design through experimentation with temperature, JSON schema prompts, chain-of-thought reasoning, and prompt chaining, establishing best practices for consistent and accurate outputs- Integrated MLflow into the experimentation pipeline to log models, prompts, and parameters, architecting connections with LangChain and other frameworks for seamless tracking and reproducibility.

Education

• R. V. College of Engineering, Bangalore

  Bachelor of Engineering (B.E.)

  2010 — 2014

• University of Maryland - Robert H. Smith School of Business

  Master’s Degree

  2016 — 2017

Skills

• Tableau
• Project Management
• Microsoft Sql Server
• Testing
• Engineering
• Machine Learning
• Microsoft Excel
• Matlab
• Program Management
• R
• Html
• Sql
• C
• Microsoft Office
• Data Mining
• Requirements Analysis
• Java
• Big Data Analytics
• Data Analysis
• Panda
• Software Development
• Python
• C++