Municipio de Guadalajara

The challenge

The Municipality of Guadalajara, through its Directorate of Innovation and Digital Services, is one of Mexico's most active local governments in digital transformation. With more than 1.5 million residents and an ambitious digital-government agenda, it set out to bring municipal procedures and services closer to citizens quickly, securely, and 24/7 — without lines or trips to government offices.

The Municipality already had the App City mobile app as its citizen interaction channel, but support for questions about procedures, tourism, cadastre, and municipal news relied on in-person or phone channels with limited hours. The challenge was threefold: embed conversational AI directly inside App City for natural-language queries in Spanish; connect in real time to municipal APIs for procedures, tourism, cadastre, news, and public calls without redesigning existing backend systems; and build a secure, scalable platform able to manage documents — generating PDFs, sending and receiving files — directly from the chat. Additionally, a RAG (Retrieval-Augmented Generation) system over official PDF documents was required to answer complex queries based on current regulations and procedures.

The solution

iNBest designed and built the Virtual Assistant for the Municipality of Guadalajara: a production-grade conversational AI agent, deployed on a Docker microservices architecture and orchestrated with LangGraph for complex agentic flows with tool-calling. The heart of the assistant is Anthropic's Claude API, selected for its contextual understanding in Spanish, its accuracy reasoning over complex regulatory documents (RAG over official PDFs), its predictable behavior in citizen-facing production, and native support for Claude Code to accelerate the development cycle.

The architecture is composed of four microservices: Chat API (the main conversation endpoint connecting citizens to the Claude engine via LangGraph), Admin API (an administration dashboard for LLM configuration, logs, and monitoring), LangGraph AI Engine (the agentic engine that orchestrates tool-calling, multi-session context management, and procedure flows), and RAG Pipeline (retrieval over the municipality's official PDF documents for precise regulatory answers).

Integrated modules include the assistant consumption API (REST + JWT/OAuth2), security and authentication (SSO + JWT integrated with App City), document send and receive (PDF, images, video), PDF management with Amazon S3 storage (signed URLs), monitoring with a logging system and dashboard, and the MCP module for municipal APIs (News/Public Calls, Tourism, and Cadastre). Key technologies: Claude API, Claude Code, LangGraph, Amazon S3, Docker, JWT/OAuth2, React Dashboard, Swagger/OpenAPI, and LangSmith for observability. The project was executed under a Scrum methodology with sprints certified through electronic signatures backed by CINCEL (NOM-151).

Results

• 4 production microservices integrated: Chat API, Admin API, LangGraph AI Engine, and RAG Pipeline
• 3 sprints completed and signed off with 100% on-time delivery (Sprints 1–3)
• 24/7 citizen access to procedures, tourism, cadastre, and news in natural, conversational Spanish
• Automatic PDF document generation directly from the chat, with no separate forms
• Assistant embedded inside App City, with no friction from additional channels
• RAG over official PDF documents for answers grounded in current regulations
• Dedicated administration dashboard to configure the LLM, review logs, and monitor in real time
• Recognized by Anthropic in the Project Glasswing program as a featured Customer-Facing AI Product Build