Santia: citizen WhatsApp bot
63,360 conversations. 9.63M messages. 7 months operating the largest citizen assistant in San Nicolás.
San Nicolás has 170,000 residents. Before Santia, handling a municipal procedure meant going in person: waiting in line, bringing copies, coming back if something was missing. Santia is the attempt to change that.
The bot operates 24/7 over WhatsApp (the channel residents already have installed and use every day), handling everything from booking a medical appointment to checking a traffic fine. It doesn't replace government: it makes it more accessible.
Architecture and services
Every conversation flows through the n8n orchestrator first, which determines how to resolve it. Informational queries (required documents, office hours, requirements) are answered via RAG against a Qdrant vector DB, fed by a document system that keeps municipal information auto-updated. Complex procedures are routed to specialized sub-agents: fines, driver's licenses, medical appointments, complaints.
Flows are classified into three types based on procedure complexity. Pure AI: the agent handles everything in natural language (queries, complaints, fines). Hybrid: AI for understanding and classification, deterministic for the critical procedure steps (license renewals, medical appointments). Pure deterministic: fixed decision trees with no LLM, for forms, payments, and external API validations. DNI and vehicle registration analysis is handled by OpenAI Vision/OCR directly in the flow.
- BuilderBotWhatsApp session handler and initial conversation routing
- GupshupWhatsApp Business Service Provider (BSP)
- n8nMain orchestrator: classifies intents and coordinates sub-agents
- QdrantVector DB for RAG: auto-updated municipal document store (source: AWS S3)
- OpenAILLM for natural language and Vision/OCR for DNI and documents
- Node agentsHigh-speed services for latency-sensitive operations. Complement n8n
- PostgreSQLPersistence for conversations, messages and escalations
- AWS S3Media storage: photos, PDFs, audio
- PuppeteerPDF generation for bills and certificates
- CRM PanelInterface for human operators handling escalations
Volume and operations
Since the mass launch in November 2025, the bot has processed over 9.6 million messages. The monthly peak was November (17,768 conversations), the launch month. The daily message average in April 2026 is ~57,000.
There's a strong weekly pattern: Sundays and Mondays concentrate the most traffic (~250 to 317 conversations/day), while Fridays and Saturdays are at the minimum (~65 to 114). The peak-to-valley ratio is approximately 4:1.
Automated resolution vs human operator
Every conversation goes through n8n, the central orchestrator. n8n classifies the intent of each message and decides the path: deterministic flows for procedures with known steps, natural language LLM responses for open queries, or escalation to a human operator when the procedure requires it.
The result: 85% is resolved without any human involvement. The remaining 15% is routed to operators for personalized attention.
- Resolved automatically: 85%
- With human operator: 15%
Findings
Juzgado and Corralón are the next automation targets
Two departments account for 55% of human escalations (Juzgado 33%, Corralón 22%). They're the clearest candidates for specialized AI sub-flows: high volume, repetitive intents, and procedures with identifiable steps. A per-department agent can cut operational load without losing the precision human operators currently bring.