Blog

RAG Evaluation - How do we actually know it's true?

The chatbot is live. The initial enthusiasm subsides. Then comes the question from management that every developer dreads: "How do we actually know for sure that the bot isn't telling the customer nonsense?"

Size matters: RAG & Data Ingestion

Size matters. At least when it comes to the data for your AI. We see it time and again in projects: The LLM (Large Language Model) is state-of-the-art (e.g. GPT-5), but the answers are inaccurate or miss the context. The culprit is often not the model, but the "chunking".
 

Optimizing RAG systems: Why reranking is the key to accurate answers

Have you ever experienced this? The RAG system provides many documents - but the answer from the LLM still remains imprecise.

State management for advanced chatbots

More than just Q&A: How chatbots manage the context of conversations (State Management)Chatbots are ubiquitous today, but their true intelligence is not just in answering individual questions, but in their ability to remember the course of a conversation.

Chatbots: the foundation of success

A chatbot is only as intelligent, precise and helpful as the data it has been trained with. This simple truth is the linchpin for the development of successful dialog-oriented AI systems. While aspects such as the user interface or integration capability are important, the content quality of the database forms the irrefutable foundation.

RAG (Retrieval-Augmented Generation) in a nutshell

RAG (Retrieval-Augmented Generation) is the way to feed external, up-to-date data into the "black box" of a Large Language Model (LLM) without having to retrain it.

Search technologies: RAG vs SolR

Retrieval Augmented Generation (RAG) is a revolutionary technology that underpins modern AI chatbots and Large Language Models (LLMs). It solves a central problem of pure LLMs: hallucination (the invention of facts) and the restriction to outdated training knowledge.

API for secure automated file uploads

A robust and secure API for uploading files is a central component for continuously and automatically supplying an AI bot with new content. This interface makes it possible to upload documents (e.g. PDFs, text files, CSVs) directly to the backend, where they are then automatically processed and imported into the chatbot's knowledge database.

AI Bot: Slack and Confluece integration

The AI Bot now has dedicated interfaces to two central platforms for teamwork: Slack and Confluence

More than just communication: How a chatbot convinces with "simple language" and JWT authentication

Chatbots automate customer service, improve efficiency and make websites more intelligent. We highlight two outstanding configurable features of the rms. AI bots: communication in "simple language" and integration using JWT authentication.

Server monitoring with Prometheus and Grafana

Prometheus and Grafana are powerful tools that help to monitor the performance and stability of websites, apps or IT systems. We use these monitoring solutions for permanent monitoring of all important system resources.

Update Debian 11 to Debian 12 (bookworm)

In June 2023 Debian 12 was released. We have tested the new distribution and already in productive use. Here is a short description of how the update from version 11 to 12 works.

Security: Fix Weak Key Exchange Algorithm in SSH

A security scan of a server reports the following result: The remote SSH server is configured to allow / support weak key exchange (KEX) algorithm(s). How can the SSH connection be secured?

AI translation in TYPO3

Translating TYPO3 pages is often annoying and always time-consuming. Here we show how content can be translated into almost any target language with a single click using deepL.

Integration of AI in TYPO3

Applications based on language models like GPT-3 or BERT are becoming more and more popular and the number of tools like ChatGPT is increasing almost daily. This article is specifically about the integration of GPT-3 in TYPO3.

Secure VPN with Wireguard and wg-easy

Automatic connection of SSH through a proxy

How do you automatically configure the use of a proxy for certain SSH connections, e.g. to use a static IP within a company network from the home office?

Content Security Policy (CSP)

Content security policy is a security feature that web designers and server administrators can use to prevent attacks such as cross-site scripting (XSS).

Shrink VirtualBox VDI Image

Virtual Box VDI images can get much relatively large over time. This is a brief guide to shrinking these images.

TYPO3 Multi-Factor Authentication

TYPO3 V11 includes 2-factor authentication in the core. This makes it easy to implement more secure authentication mechanisms for all backend logins.

Thumbnails and pagination in TYPO3 11

Missing thumbnails in the backend and no longer working pagination for news or solr in the frontend. This small tutorial shows how to display the thumbnails again.

Translate XMLs with DeepL

Automate complex and laborious manual translation processes using AI from DeepL via a php API.

TYPO3 PageTitleProvider for own extensions

Creating custom PageTitleProviders for TYPO3. Use cases are extensions that use a single page for displaying different content like single views of news or blog posts.