Visit to Morocco · Public documentation

From conversation to business action.

This documentation is based on the project README, four n8n screenshots and a demonstration video that was inspected but intentionally not published.

Summary

Visit to Morocco is an experimental travel agent available through Telegram and orchestrated with n8n. It analyzes traveler requests, selects tools and prepares information for human sales follow-up.

Status: Functional prototype in a development environment.
Type: Personal project for a fictional agency.
Executions: Several successful executions visible on August 20, 2026.
Production: No public deployment or client use claimed.

Main architecture

Telegram Trigger
      ↓
OpenAI AI Agent ← conversation memory
      ├── Vector Store + embeddings
      ├── SerpAPI web search
      ├── Gmail
      ├── Google Sheets
      └── Google Calendar
      ↓
Telegram response

The model makes decisions and generates responses. n8n connectors execute external actions. This separation makes the system more than a text-only chatbot.

RAG document pipeline

Google Drive Trigger
      ↓
File download
      ↓
Data Loader → Recursive Text Splitter
      ↓
OpenAI Embeddings
      ↓
Simple Vector Store

The pipeline turns documents added to Google Drive into searchable chunks. Simple Vector Store suits experimentation; the README plans a move to PostgreSQL and pgvector for stronger persistence.

Business actions

Progressively qualify destination, duration, dates, travelers, interests, budget and contact details.
Consult a document base for information belonging to the fictional agency.
Use external search for potentially recent public information.
Record a lead in Google Sheets.
Send a communication through Gmail.
Google Calendar is configured for appointments; its activation is not shown in the published evidence.

Evidence level

Demonstrated: Telegram conversation, context memory, qualification, response, email sending, Google Sheets writing and successful n8n executions.
Visible configuration: OpenAI, Vector Store, embeddings, SerpAPI, Gmail, Sheets, Calendar and the Google Drive pipeline.
Not claimed: Production use, client results, large-scale reliability, continuous availability or measured time savings.

Potential business impact

The system can absorb part of first-line work: recurring questions, information collection, research, data entry and confirmation. A consultant can receive a more complete file instead of starting from an unqualified message.

The project-page simulator uses requests × manual minutes × automatable share ÷ 60. It is an adjustable projection tool, not a results study.

Limits and next steps

Simple Vector Store is not suitable for persistent production use.
Edge-case tests and a human escalation policy must be formalized.
Permissions, logging and error handling need reinforcement before real use.
The bot and screenshots contain spelling variants of the project name that need standardization.
The raw video contains test contact details and remains excluded.