-
1. Talks Understands what the prospect wants —an AI agent or a training programme— in Spanish or English.
-
2. Gathers requirements Asks for what is missing, with clickable options, and stores each answer in its own field.
-
3. Verifies the email Sends a 6-digit code and validates it before moving on.
-
4. Sends the PDF brief Generates the requirements brief and emails it to the verified address, copying our team.
The challenge
Every enquiry reaching the Academy —through the website, WhatsApp or email— started the same way: several back-and-forths to understand what the prospect needed, which systems they used, how many people were involved or what they expected to invest. Much of the information arrived incomplete, some got lost in the chat, and the sales team spent hours on repeated questions.
On top of that, anyone could leave a mistyped or made-up email, and without an organised record there was no way to prioritise who to call first or prepare a proposal based on reliable data.
The solution
We built a conversational AI agent that now answers on academiainteligenciaartificial.com, both in the floating chat and on the full chat page. It replies in real time (streaming) and in the prospect’s language, Spanish or English, thanks to a dedicated language classifier.
Several components work in every turn: a planner decides which data the prospect provided and stores it with one tool per field; a responder writes the reply and the next question. The critical decisions are made by code, not by the model: which field to ask next, which tools are available for each business line and when the brief is ready to close.
It first identifies whether the prospect wants to automate a process with an AI agent or train their team. For automation it records the process to automate, the type of agent, current systems, scope, team size, industry and budget expectation; for training, the technical level, type of course, objective, participants and format. When useful, it shows single- or multiple-choice questions. Budget is recorded as qualification data: the agent never quotes.
Then it asks for name, phone, company and email, and verifies the email with a 6-digit code that expires after 10 minutes and allows 3 attempts; only its hash is stored. Once everything is complete and the email is verified, it generates the Requirements Brief as a PDF —in the language of the conversation, with the details exactly as the prospect stated them and protected against editing with AES-256 encryption— and emails it to the verified address, with a copy for our team.
Every message first goes through Azure AI Content Safety, the service is protected with an API key, CORS and rate limiting, conversations are stored in PostgreSQL and every turn is traced with OpenTelemetry. The agent’s prompts were optimised with automatic evaluation on separate training and test sets of conversations.
Results
- 24/7 sales conversations on the website, in Spanish and English.
- Every lead arrives with a structured brief and a verified email, ready for the proposal.
- The prospect gets a formal document with their requirements without waiting for a meeting.
- Fewer back-and-forths for the sales team and full traceability of every conversation.
- It runs in production: you can try it now and receive your own PDF brief.