Pacte Xat

Español: For the Spanish version of this README, see README.

Overview

This repository defines Pacte, a full-stack AI-driven chat platform built to empower local governments and institutions with secure, data-driven decision-making capabilities. Using regional data, Pacte enables planning and decisions as good as—or better than—central governments and supranational organizations like the WHO or UN. The system is powered by a Node.js Fastify server, a cross-platform chat interface (web and native mobile via Expo), a LangChain orchestrator for multi-agent workflows, and data crawlers to gather and process regional information.

Key components include:

1. Chat Interface: A ChatGPT-like experience for users to query data, generate insights, or request actions. Ask questions like "What’s the health trend in my region?" or "Plan a flood response."
2. Data Crawlers: Collect and structure regional data for analysis.
3. Agentic Workflows: Multi-agent system (e.g., Data Agent, Analysis Agent, Planning Agent) orchestrated via LangChain.

Primary Goals

1. Empower Local Decision-Making

   • Provide local governments with tools to analyze regional data and plan effectively.
   • Outperform centralized systems by focusing on hyper-local context.

2. Multi-Agent Coordination

   • Leverage LangChain to orchestrate agents for tasks like data crawling, summarization, and decision support.
   • Central Planning Agent coordinates sub-tasks; specialized agents handle data and analysis.

Core Components

1. Fastify/Node.js Backend

   • TypeScript-based server with APIs to manage chat sessions, process data, and execute agent tasks.
   • Integrates with containerized environments for secure computation.
   • Serves the chat interface and mobile app backend.

2. Cross-Platform Chat (Expo)

   • Built with Expo for web and native mobile (iOS/Android) compatibility.
   • Provides a ChatGPT-like UI for querying data and interacting with agents.

3. LangChain.js Orchestrator

   • Manages multi-agent workflows (e.g., crawling data, generating reports, suggesting plans).
   • Powers LLM-driven responses and task delegation.

4. Data Crawlers

   • Custom scripts to scrape, aggregate, and clean regional data from various sources.
   • Feeds structured data into the system for agent analysis.

ToDo

• Bare minimum web and app working on Expo
• Add LangChain agents to the server
• Add UI for à la ChatGPT
• Scrape data from some local government websites and regions
• Put scraped data into a database