Bringing drag-and-drop simplicity to AI development

The client needed to develop custom AI assistants, but the process was complex, costly, and required specialized AI teams. Adapting models to unique knowledge bases involved demanding RAG workflows, API calls, and data management, which slowed deployment and limited scalability. Without a solution, the client risked losing efficiency and competitive advantage in customer support and communication.

The client was provided with the AddAI platform, which leverages Llama models to turn AI assistant development into a drag-and-drop tool without the need for programming. The solution automates training, fine-tuning, pipeline orchestration, and RAG workflows, and includes a wizard that guides users from prompt creation to deployment. With a high degree of automation, even non-technical teams can quickly deliver and manage their own assistants, significantly accelerating implementation and reducing costs.

The main users of the solution are non-technical business and customer teams, who independently manage AI assistants for client support. Outputs are primarily used by customer service staff, contact centers, marketing, and communications, where assistants provide quick and accurate responses to end customers. Product and business managers also use the platform for easy content tuning and conversation personalization without the need for IT or data specialists.

LLM models: Llama 3.3 70B, Llama 3.2 11B, Llama 3.2 3B, Llama 3.1 405B. Platform: IBM watsonx (for development, fine-tuning, pipeline orchestration). Technological approaches: Retrieval Augmented Generation (RAG) workflows, multi-model orchestration, drag-and-drop no-code framework.

Head of Innovation / Digital Transformation

Joint implementation with the client

Hybrid delivery with vendor and internal team.

Customer service, sales and marketing teams, product and business managers, and system specialists.

Ideally cloud infrastructure for running LLM models and orchestration.

Customer communication quality improved thanks to more accurate and faster responses, with unanswered queries reduced by more than 50%. Support processes became more efficient, saving thousands of staff hours monthly, and user experience improved with consistent and personalized service.

500,000+ customer interactions per month,  85%+ accuracy of AI assistant responses, 50%+ reduction in unanswered queries, 6,000 staff hours saved every month.

Key factors included the deployment of LLM models, the intuitive AddAI no-code platform, solid data foundations for RAG workflows, and close collaboration between the vendor and user teams.

The biggest challenge was the complexity of developing and customizing AI assistants for the client’s unique knowledge bases. We overcame this by delivering a no-code platform with automated RAG workflows and a wizard that simplified the entire process even for non-technical teams.

We recommend choosing open and flexible technologies, preparing data sources thoroughly, and ensuring user-friendly operations for non-technical staff. It is advisable to involve non-technical teams in the project.