DACHNUG

A few years ago, chatbots had very limited skills relying on predefined answer patterns, frustrating users and developers alike. This drastically changed with the release of the GPT 3. It’s enhanced language understanding and generation abilities changed the ceiling of what a chatbot can do and what not. But the open source AI market didn’t sleep on those developments and quickly build models even more powerful than what we know from the early days of ChatGPT - making building solutions with LLMs even more accesible.
One challenge that remained when building a useful company chatbot was the addition of your own data into the mix - especially if it wasn’t previously puplicly available (like internal company guidelines) or data that changes quickly (press releases, incident messages). Retrieval Augmented Generation introduced a framework to tackle this problem by smartly using the power of Vector Databases and Semantic Embeddings.
The release of powerful development kits reduce the complexity of building a chatbot even further, allowing even inexperienced AI developers to set up their frist working agent in little time - creating a world in which building your first chatbot as easier than ever.

In my talk I want to explain the fundamental concepts of building a company chatbot with large language models, Retrieval Augmented Generation, Vector Databases, Semantic Embedding and AI Function Calling while simultaneously using these concept to implement a simple chatbot using open source LLM models from german hosted IONOS AI Model Hub. By attending, the audience should be enabled to start building AI chatbots themselves and understand the core ideas behind it.