Project: QChat

A conversational agentic platform designed to make financial literacy accessible, actionable, and engaging for youngstars. It explains complex financial concepts in clear, jargon-free language, provides real-life examples with an investment context, and delivers answers in structured formats.

Active Features

• Streaming Responses(SSE): Engage in conversations with LLMs with Chat renders token-by-token from /chat-stream for an interactive user experience.
• Agent Tool Telemetry: Inline stages show what the agent is doing—searching → reading → writing—with tool outputs surfaced where helpful.
• Generative UI-First Rendering: Renders JSON UI specs (charts, tables, KPIs) generated by the agent via C1 UI generation—powered in the app by @thesysai/genui-sdk.
• Web Search Integration: The Agent can perform web searches (web crawl-based) on demand to provide up-to-date and relevant information, displaying concurrent results to the user.

See the presentation and demo on YouTube: Click here

System Architecture

The server's workflow is orchestrated by a LangGraph state machine:

• Request Handling: A user query is sent to the /api/v2/chat-stream endpoint.
• Graph Initialization: A new / existing conversational graph is configured. A system prompt (CUSTOM_SYSTEM_PROMPT) is injected, instructing the agent to act based on it.
• Model Invocation: The agent's core LLM make decision the next step: respond directly or use a tool.
• Tool Routing: If a tool is chosen, the graph routes the state to the tool_node.
• Tool Execution: The selection tool (Tavily, YahooFinanceNews, or c1_ui_generate) is executed.
• Response Generation: The output from the tool is passed back to the model, which decides whether it is the final response.
• Streaming: Throughout this process, events (content, tool_starts, tool_outputs, end) are streamed back to the client as SSE.

PS: This project originally was a part of a hackathon and is still in its early stage; Star the project on GitHub to follow/contribute alongside ✌️