Report Abuse

Basic Information

This repository is a personal learning project for building Agentic AI systems using the OpenAI Agents SDK and demonstrates an interactive financial analytics application focused on IDX-listed companies. It showcases modular, LLM-driven workflows where an orchestrator agent parses user intent and coordinates specialized sub-agents such as a company overview agent and a trend analyst agent. The app integrates retrieval-augmented generation (RAG) to fetch live financial data from the Sectors.app API so analyses and visualizations reflect current market data rather than model hallucination. The project is implemented with Streamlit for a web interface and Plotly for charts, and it emphasizes practical agent patterns, input validation with Pydantic, and guardrails to restrict queries to supported companies and compliant topics.

Links

Categorization

App Details

Features
The repository documents and implements multiple agentic patterns and practical features. It uses the OpenAI Agents SDK for multi-agent coordination, includes function-based tools to call the Sectors.app API for real-time IDX data, and provides RAG-enabled analysis. Visualization is handled with Plotly and outputs are integrated into Streamlit for interactive exploration. Architectural features include a triage agent and an orchestrator agent, an agent-as-tool approach, a Planner–Executor pattern for multi-step queries, and Pydantic-based schema validation for agent outputs. The app also implements input guardrails to block unsupported or non-compliant queries and a chat mode that maintains conversation state while delegating company-related questions to the orchestrator.
Use Cases
This project is useful as a hands-on reference for developers and researchers exploring multi-agent LLM applications and RAG in the financial domain. It provides concrete examples of orchestrating multiple tool-agents, enforcing strict input and output schemas with Pydantic, and preventing hallucination by calling live APIs. The documented learning milestones illustrate how to evolve architectures from simple triage to agent-as-tool and Planner–Executor patterns, which improves reliability for multi-step queries. The Streamlit demo and chat mode show how to present conversational, multi-turn interactions with preserved structured outputs and visualizations. Input guardrails demonstrate practical safety and scope-limiting techniques for production-readiness.

Please fill the required fields*