Agentic AI and IBM watsonx Orchestrate Usage - 2026 hackathon watsonx challenge

Описание: Our solution uses IBM watsonx Orchestrate as the central orchestration layer with a hierarchical multi-agent architecture. We built a supervisor agent called BAU Enhancement Estimate Agent in watsonx Orchestrate, which uses three Langflow agents as tools: Code Agent, Enhancement Agent, and Estimation Agent. We also built an Email Rewriter Agent directly in watsonx Orchestrate.

All Langflow agents implement RAG architecture with data stored in IBM Cloudant vector databases. We have three specialized databases for code patterns, enhancement strategies, and estimation templates. The knowledge base documents were processed using Docling for intelligent chunking before being embedded and stored in Cloudant.

When a user submits a request through our React frontend deployed on IBM Code Engine, the supervisor agent analyzes the intent and routes it to the appropriate specialist agents. For complex requests requiring both code analysis and estimation, it orchestrates multiple agents sequentially. Each specialist agent retrieves relevant context from Cloudant and generates responses using IBM Granite models via watsonx.ai.

The supervisor agent then aggregates responses from sub-agents and presents a unified answer to the user. The Email Rewriter Agent operates independently, using IBM Granite to transform informal drafts into professional communications.

This architecture demonstrates the full power of IBM's agentic AI ecosystem—watsonx Orchestrate for orchestration, Granite models for intelligence, Cloudant for RAG storage, Langflow for visual agent building, and Code Engine for application deployment.