Naisarg Halvadiya

About

Built a production agentic AI assistant in Python using Playwright plus Google Gemini’s vision capabilities to automatically document and execute UI workflows step-by-step, reducing developer time spent on trivial documentation/knowledge transfer. Also built an Apache Airflow ETL pipeline and has experience evaluating AI agents with human-in-the-loop methods, plus successfully communicated a vision-model-based CMS analytics PoC to non-technical university stakeholders and proposed it to Academic Technology with cost-savings rationale.

Experience

Education

Key Strengths

• Built and deployed an agentic AI system that automates step-by-step UI workflows using screenshots + DOM events
• Practical integration of Playwright with an LLM vision model to interpret UI state and drive actions
• Solved UI automation reliability issues by tracking DOM structure changes to avoid loops and improve next-action selection
• Hands-on orchestration experience building an Airflow ETL pipeline from source ingestion through transformation to database load
• Iterative prompt refinement approach (starting zero-shot, adding context/instructions based on output deviation)
• Human-in-the-loop evaluation methods for agent reliability (step documentation + human classification / real-time observation)
• Able to translate complex AI/vision concepts to non-technical stakeholders and drive buy-in for a successful PoC