Published on 04/07/25

Our chatbot that doesn't just respond

Arthur
Arthur Guillermin Hazan
3 min
Share:

A conversational assistant that understands... and suggests

Asking a question and getting a clear answer is great. But when the answer is accompanied by direct access to the right content, it's even better. This is exactly what the chatbot module developed for our site enables.

Designed as a real navigation assistant, this chatbot doesn't just answer: it guides, suggests and enriches the exchange by proposing related content, in an interface designed to enhance exploration.

How does it work?

  • Open via the main menu

    Accessible from the site menu, the chatbot opens with a selection of suggested questions. Users are also free to formulate their own requests.

  • Question analysis

    The AI relies on a structured and mastered corpus:

    - site content (over 100 active pages),

    - internal documents (presentations, commitments, operations),

    - pre-constructed answers to the most frequently asked questions.

  • Answer generation

    AI generates a clear, concise and autonomous answer, intended to provide a complete first response to the question posed.

  • Contextual enrichment

    Once the response has been generated, the AI searches our vector database for the content closest to the subject at hand (articles, case studies, expert reports, etc.). These elements enrich the response in the form of visual blocks integrated into the thread of the discussion.

AI Module

Integrated suggestion modules

Several blocks visually enrich the chatbot's responses, transforming a simple text response into a guided tour:

  • Case study" block: carousel of case studies related to the question
  • Insight" block: selection of related articles
  • Expertise" block: highlights relevant expertise
  • Single link: action button to specific content

Our approach

These modules are triggered automatically, depending on the nature of the question and the proximity of available content. This system overcomes the limitations of conventional chatbots by adding a "suggestion" dimension often absent in conventional approaches.
 

 

Langfuse

Observing for better adjustment

To monitor and improve the chatbot's effectiveness, an observability solution has been implemented via Langfuse. This makes it possible to :

  • visualize the questions asked and the answers generated,
  • identify friction points or areas of confusion,
  • refine prompts and sources as usage progresses.

Our approach

This progressive approach guarantees continuous improvement in the relevance of answers, while keeping the technical impact under control.

 

Our AI on the site

We open the doors of our AI to you with complete transparency: what it does, what it doesn't do.

What about you?

Interested?
Contact

This module has been designed to be deployable in other contexts.

If you'd like to integrate an enriched chatbot into your own digital ecosystem, let's talk!

These articles may be of interest to you

Would you like to talk about it?

Contact us!
Contact

According to my analysis

Are you interested in our projects and our expertise? Why don’t we have a chat?