Chatbot and recommendation engine development

01 The Challenge

Our Client needed a more stable and capable chatbot that could hold human-like conversations and support customer onboarding and product recommendations, all within the tight technical confines of WhatsApp.

Their goal was to build a reliable support channel accessible even in areas with low Internet connectivity – one that could onboard new clients, answer questions about available products, and provide personalised suggestions based on user data.

When Altamira joined, the project was still in its early stages of development. The codebase was limited in scope and poorly structured. The Client was looking for a long-term development partner to rethink the architecture and gradually evolve the chatbot into a stable, AI-enhanced assistant and recommendation engine that could scale.

Together with the Client’s internal team, we defined the development stages, starting from rewriting the bot in .NET, moving on to AI integration, and ending with a customised recommendation engine.

From day one, we brought a focused product mindset to the table: planning each stage, recommending the best-fitting technologies, and advising on feasibility and rollout risks. At the same time, our Client was heavily involved during key decision points, especially around data privacy and AI integration.

The most significant technical challenge came during the implementation of AI-based responses using Snowflake. Although the data was clean and well-prepared, Snowflake’s performance under high load resulted in performance issues. We quickly pivoted, transitioning to AWS Bedrock for a more stable and scalable AI foundation.

02 The Solution

We started by using WhatsApp’s API in ways most don’t. Instead of a static conversation, we crafted an interactive onboarding journey. The bot first asked for the user’s geolocation, which is standard for WhatsApp, but we used it to autofill the user’s address in a registration form. On top of it, we used advanced WhatsApp features like Flows, which allow the development of complex data input forms within the messenger, similar to web forms. From there, the user would share their name, email, and password, without ever leaving the chat.

Once registered, new users were invited to explore the company’s services or continue the conversation. If the user had been with the company before, we’d recognise their phone number and skip onboarding, moving straight to personalised service recommendations.

The turning point came when we began building the bot’s brain.

The Client had compiled a document packed with details – services, markets, product offerings, core company goals. We used that to train a RAG (Retrieval-Augmented Generation) engine, creating a private, internal knowledge base that the bot could draw from to answer real questions. This wasn’t menu-based navigation but a natural conversation, like asking a well-trained team member.

Additionally, besides providing information via RAG, the AI agent provides capabilities for handling any queries related to the shared information, including making calls to external services.

Each conversation now followed a different flow, depending on who the user was and what they needed. Behind the scenes, we orchestrated three major integrations: the Client’s own services platform, the recommendation engine, and AWS’s AI tools.

03 The Result

What began as a basic, one-way bot has evolved into a smart assistant that can understand context, guide users, and surface relevant offers, all within WhatsApp, with no downloads or browser required.

• New users onboarding time dropped by 83%, thanks to the smooth and clear WhatsApp-based registration flow with prefilled data.
• The AI-powered FAQ, trained with the company’s own documentation, led to a 60% reduction in support tickets, easing pressure on the support team.
• Personalised flows and the integration of the recommendation engine resulted in a 3x increase in engagement among returning users.
• The .NET rebuild gave the system a solid foundation – stable, scalable, and secure.
• RAG-enhanced answers provided the bot with the depth to respond with context and clarity.
• The recommendation engine integration brought personalisation into the picture, turning a basic chat into a real sales assistant.
• Migrating to AWS Bedrock gave us performance at scale.

Today, the bot doesn’t just answer questions but builds relationships. And most importantly, our Client now has a digital assistant that reflects their brand in every conversation: efficient, helpful, and human-like.