Conversation with memory
Telegram, an OpenAI model and memory preserve context.
Project 03 · Business AI agent
Visit to Morocco
A travel agent that turns a Telegram conversation into a structured commercial journey: understand, search, recommend, qualify and hand over to the business.
STATUSFunctional prototype in development
TYPEFictional business demonstration
EVIDENCE4 verified screenshots
CLIENTNone — personal project
01 — Business challenge
One incoming request should not create five manual tasks.
In an agency, answering the traveler is only the beginning. The need must be understood, a reliable offer researched, contact details collected, the lead entered, confirmation sent and sometimes an appointment scheduled. Visit to Morocco orchestrates this chain from one conversation.
MONTHLY SCENARIO100
requests to process
AUTOMATABLE SHARE70%
adjustable simulator assumption
POTENTIAL TIME14h
of repetitive work reallocated each month
Development-environment estimate: 100 requests × 12 minutes × 70%. This is not a measured client result.
02 — System effect
A message enters.
A usable file comes out.
The agent does not replace the consultant. It prepares the ground so the person intervenes when the request becomes commercially relevant.
- 01INPUTThe traveler writes in Telegram
A familiar, available interface.
- 02UNDERSTANDINGThe agent qualifies the need
Destination, dates, duration, travelers, interests and budget.
- 03RELIABILITYThe system checks its sources
Document base for internal offers, web search for external information.
- 04ACTIONTools take over
Sheets record, confirmation email and configured calendar.
- 05HUMANThe consultant receives a prepared request
Less re-entry and more time to advise and close.
03 — What is built
More than a chatbot: an operational chain.
Document RAG
Google Drive, extraction, chunking, embeddings and Vector Store search.
External search
SerpAPI is available when an answer depends on recent public information.
Commercial qualification
The dialogue progressively collects trip, budget and contact details.
Automated handover
Google Sheets recording and email sending are confirmed by the demonstration.
Human handoff
Google Calendar is configured to schedule a consultant discussion.
04 — Verifiable evidence
The architecture is visible.
So are the executions.
Original n8n workflow screenshots captured on August 20, 2026. No video frame containing contact details is published.
Main workflow
The agent receives the Telegram request and has seven components or business tools.
RAG document pipeline
A Drive file can be prepared and added to the vector base.
Successful execution
The Telegram → agent → response path finishes successfully.
Execution history
Several successful development trials are visible.
05 — Simulate the potential
How much time could this chain release?
Adjust assumptions to your business. The calculation estimates manual time potentially absorbed by automation.
MONTHLY POTENTIAL14 h
of repetitive work reallocated
100 requests × 12 min × 70%
Indicative simulation, not a measured client result.06 — Technologies
A connected-agent stack.
ORCHESTRATION & INTERFACE
n8n · Telegram Bot
INTELLIGENCE & KNOWLEDGE
OpenAI · Embeddings · RAG · Simple Vector Store · conversation memory
BUSINESS TOOLS
Google Drive · Gmail · Google Sheets · Google Calendar · SerpAPI
Transparency
Visit to Morocco is a personal project and the agency is fictional. Screenshots prove workflow configuration and several successful development executions; they are neither client results nor production proof. The raw video remains excluded to protect visible contact details.