Tejas Haral

EY Global Delivery Services (EY GDS) · Pune, India

• Advised cross-functional delivery teams on NL-to-SQL architecture, prompt and context engineering strategy, and agentic AI feasibility, acting as the technical point of contact for GenAI design decisions.
• Architected and shipped 2 Azure AI Foundry agent-based GenAI POCs end-to-end — from first line of code to a working deployment — using asynchronous Python/FastAPI services, LangChain/LangGraph orchestration, and GitHub Actions CI/CD on a single-server environment.
• Designed multi-agent orchestration workflows on Azure AI Foundry and LangGraph with agent harnesses, guardrails, and Responsible AI controls to enable safe, autonomous task execution for enterprise GenAI use cases.
• Built POC application infrastructure on Azure Cosmos DB, Azure App Service, and Azure Static Web Apps, covering data persistence, hosting, and end-to-end integration for GenAI proofs of concept.
• Fine-tuned GPT-4.1 in Azure AI Foundry on custom synthetic datasets and prototyped agent behavior tuning to adapt LLM responses for domain-specific enterprise tasks.

KPIT Technologies · Pune, India

• Engineered and deployed generative AI modules using RAG and prompt engineering to automate test case and test script generation/execution, cutting manual effort by 40%.
• Built intelligent document parsing pipelines with OCR and OpenCV, converting unstructured enterprise documents into ML-ready inputs and cutting preprocessing time by ~35%.
• Embedded generative AI techniques into legacy enterprise software to reduce process latency and improve software efficiency by 30%, while maintaining production readiness.
• Architected multi-agent AI systems with retrieval tools to automate document understanding, reducing manual document processing time by ~25% across cross-functional teams.
• Trained a YOLO object detection model on car infotainment icons and UI elements, boosting detection accuracy by 50% and enabling automated visual regression testing.
• Engineered a multi-object road detection Deep Neural Network for ADAS safety features, replacing manual image annotation with an automated pipeline and improving model throughput.