AI agents
AI Reception 24h: an AI receptionist that answers around the clock
An AI-powered reception for centres and offices with 24-hour access. It answers by phone and WhatsApp, recognises the caller, verifies their identity and, when someone cannot get in, handles opening the door at any hour.
Results
- Support by phone and WhatsApp
- 24/7
- Languages: Spanish and Catalan
- 2
- Verification: phone number and one-time code
- 2 steps
- Conversations logged and auditable
- 100%
01
The challenge
In a centre or office with 24-hour access, users need help at any time, but staff are only on site a few hours a day. Out of hours, a credential that fails to read leaves someone at the door with nobody to call, and questions about prices, opening hours or services pile up until the next day. The centre needed a reception that answered instantly, by phone and in writing, in Spanish and Catalan, and knew who it was talking to before acting. With one condition: artificial intelligence could not have the final say on opening a door. It had to be helpful and secure at the same time.
02
The solution
We designed and deployed an AI agent that works as a virtual receptionist and integrates with the centre’s systems: telephony, WhatsApp Business, the user database, access control and incident management. Everything is supervised from an operations dashboard.
- Answers calls and WhatsApp messages 24/7 with a natural voice, in Spanish and Catalan, with no waiting and no option menus.
- Recognises the user by phone number and confirms their identity with a one-time code sent by WhatsApp.
- Diagnoses the access problem with simple questions: whether the credential is not reading, whether the membership is active or whether the reader is faulty.
- Separates conversation from decision. The AI only requests the opening; an independent rules engine checks identity, time, centre and user status and sends the command to the door via API. When in doubt, the door stays closed.
- Hands cases that do not fit to a person. The team reviews them in the dashboard and decides with one click whether to open.
- Logs every conversation with its steps, its cost and a tamper-proof audit trail, and opens an incident automatically when it detects a fault.
How it handles a call, step by step
- 01
Takes the call or WhatsApp
The voice agent answers on the first ring, at any hour, and detects whether the user speaks Spanish or Catalan.
- 02
Identifies the caller
Looks up the phone number in the user database and retrieves their centre, membership and status.
- 03
Verifies identity
Sends a one-time code by WhatsApp and checks it before taking any action on access.
- 04
Diagnoses the problem
Asks what the reader or app shows and tells apart an unread credential, an inactive membership or a hardware fault.
- 05
Requests the opening
Passes the request to the rules engine, which validates identity, time, centre and status and only then sends the command to the door.
- 06
Logs and escalates
Stores the conversation and every decision in the dashboard, opens an incident for faults and hands doubtful cases to the team.
Security: the AI talks, the rules decide
The main risk of an AI agent connected to physical systems is that a well-crafted conversation persuades it to do something it should not. That is why the architecture separates two parts: the language model understands the user and proposes an action, and a deterministic rules engine, out of the conversation’s reach, is the only component that can order the door to open.
The engine requires an identity verified with a one-time code, an allowed time, the right centre and an active user. If any condition is missing, the default is to keep the door closed and pass the case to a person. Every step is stored in a tamper-proof audit trail the team can review in the dashboard.
AI agent integrations
- 01
Telephony and real-time voice
Speech recognition and synthesis for a natural phone conversation, with no option menus.
- 02
WhatsApp Business
Written channel for questions and for sending one-time verification codes.
- 03
User database
PostgreSQL lookups of memberships, assigned centre and user status.
- 04
Access control via API
The rules engine sends the opening command to the door of the right centre.
- 05
Incident management
Faults detected during the conversation are logged as incidents automatically.
- 06
Operations dashboard
Conversations, tools used, verification status, cost and manual approval of openings.
What artificial intelligence brings to the business
The centre offers 24/7 support without adding staff. Questions about prices, opening hours and services are answered on the spot, and out-of-hours access problems are handled with the same safeguards as if someone were at the front desk.
The team stops taking repetitive calls and focuses on cases that need judgement, which reach them already identified and documented. Management gets full traceability: what was asked, what the system decided, why, and what each conversation cost.
The same agent model works for coworking spaces, business centres, 24-hour gyms, self-storage, car parks or any company that has to serve its users outside office hours.
Technologies
TypeScript
React
PostgreSQL- AI voice agent
WhatsApp Business API- Access-control API
Process
- 01
Discovery
Analysis of frequent questions, access problems and each centre’s opening rules.
- 02
Agent design
Conversation in Spanish and Catalan, agent tools, rules engine and criteria for handing over to a person.
- 03
Integration
Connection to telephony, WhatsApp Business, the user database, access control and incidents.
- 04
Production and supervision
Testing with real calls, operations dashboard and continuous improvement from logged conversations.
Gallery
Swipe to see more · tap to enlarge
FAQ
Frequently asked questions
It is an AI agent that handles calls and messages the way a front-desk person would: it answers questions, identifies who is getting in touch and carries out tasks by connecting to the company’s systems. Unlike an answering machine or an option menu, it holds a natural conversation and resolves the request on the spot.
Yes, as long as the AI is not the one making the decision. In AI Reception 24h the agent only requests the opening; an independent rules engine checks the code-verified identity, the time, the centre and the user’s status. If anything does not match, the door stays closed and the case goes to a person.
Telephony, WhatsApp Business, the user database, access control via API and incident management. In other projects we connect AI agents to CRMs, ERPs, calendars and phone systems: the integration is defined in the initial audit based on the systems the company already uses.
This project serves users in Spanish and Catalan and detects the language automatically. The agent can be configured for other languages depending on each company’s audience.
It hands it to the team with the conversation, user identification and diagnosis already done. From the dashboard, a person reviews the case and decides with one click, without asking the user again.
Any company that has to serve users or customers outside office hours: coworking spaces, business centres, 24-hour gyms, self-storage, car parks, residences or field-service companies. The first agent is usually in production in about 8 weeks.
Want an AI agent like this in your company?