*This post was created with a self-made CustomGPT based on Sophie’s latest YouTube video. You can find the link to the video at the end of this article.
Rasa is back – with a revised concept that focuses on the use of Large Language Models (LLMs). In this podcast episode of Sophie’s Next AI Talk, Sophie talks to Sebastian from Rasa about the company’s technological transformation, the new hybrid approach called CALM and what this means for the future of Conversational AI.
A personal comeback: why Rasa plays a special role for Sophie
For Sophie, this podcast episode is more than just a technical update. Her own AI journey began with Rasa – back then in one of Switzerland’s first chatbot projects. This makes it all the more exciting to hear how the former open source start-up has developed and is now back on the Conversational AI stage.
The turning point: from LLM skeptic to CALM framework
For a long time, Rasa stayed away from LLMs like ChatGPT. While other providers quickly jumped on the bandwagon of the new technology, Rasa continued to rely on classic NLU systems. But that has changed. With the CALM framework (Conversational AI with Large Language Models), Rasa is now pursuing a hybrid approach that combines the best of both worlds:
- Structured business flows for maximum control
- LLMs for language enrichment for more natural conversation
- Flexible choice of model: OpenAI, Mistral, LLaMA & Co. can be integrated depending on the use case
- On-premise deployment for maximum data sovereignty
Don’t Prompt and Pray – control instead of black box
One of Rasa’s key arguments is that companies need control over their chatbots. “Prompt and pray” – i.e. blindly relying on the answers of an LLM – is not a sustainable solution in Rasa’s view. With the CALM approach, companies retain control over their responses at all times and can deploy LLMs in a targeted manner, e.g. for specific target groups, channels or use cases.
Happy Path + LLM = less effort, more quality
Even if Rasa is not a plug-and-play solution, the new approach significantly reduces the development effort. Companies define a so-called “happy path”, i.e. the ideal course of the conversation – exceptions and special cases are then handled by the LLM. This saves up to 80 % of development time compared to previous solutions.
No-Code meets Pro-Code: One system for all teams
Another advance: Rasa now offers a graphical user interface for non-programmers. The pure bot dialog can be created in a low-code / no-code environment, while the technical integration is handled by a small team of experts. This means that even specialist departments without in-depth technical expertise can design their own chatbots – including API connection, channel control and automation.
Data protection, scalability and voice
Rasa offers a real advantage, especially for data-sensitive sectors such as banking or insurance: complete on-premise hosting. Customer data remains under the company’s control – ideal for use in highly regulated environments. Voice applications can also be realized with suitable models, such as fine-tuned LLaMA models with extremely low latency.
Conclusion: Rasa has learned – and now delivers
Rasa was quiet for a long time – but definitely not idle. With the CALM framework, the company now provides a highly flexible, secure and scalable solution for conversational AI that has nothing to hide from the big players. For companies that want to improve their customer experience with AI without losing control, Rasa could be the right partner!
Any further questions?
Do you have any questions? I am happy to support you, act as a sparring partner and answer your questions. I am always happy to receive your messages, preferably by WhatsApp message or e-mail.
