Manan Suthar

Developed a sophisticated multi-agent orchestration system using Google ADK and Python to provide strategic intelligence on customer demand, partner capacity, and sales pipeline. Designed and implemented three specialized sub-agents for data retrieval with intelligent routing and query decomposition logic. Deployed the agentic solution on Vertex AI Agent Engine and Cloud Run. Engineered advanced prompt engineering techniques including ReAct (Reasoning + Acting) framework. Contributed to evolving the solution architecture from a custom RAG model to an Agentic AI solution. Built the core AI agent using Google ADK, Gemini modal and a FastAPI stream API. Owned the project extension phase, deploying new vector indexes and achieving 92% semantic similarity. Built and deployed production microservices for SQL validation, execution, and natural language answer generation using SQLGlot, Fast API, Google Gemini, GKE and Snowlflake APIs.

Created ground truth data for medical information and developed an automated testing pipeline using Python to evaluate LLM extraction accuracy to reduce human effort of manual evaluation. Applied prompt engineering (Few-shot, CoT) to improve data extraction and achieved more than 75% of accuracy.