Blog

Maximum flexibility, zero risk: how to securely integrate AI chatbots into your IT infrastructure

The introduction of artificial intelligence in companies often fails not because of "if", but because of "how". In particular, when it comes to chatbots that process customer or employee data, the alarm bells are ringing for IT managers and data protection officers. The fear of GDPR breaches or security gaps is great - and justified.

No more "one-size-fits-all": the architecture behind true chatbot personalization

Why generic answers are a thing of the past and what technical foundation is needed to dynamically tailor AI content to the user, role and context.

AI on the website: What about the GDPR?

Why data protection is not a brake on innovation, but your competitive advantage. "We would like to use AI, but our legal department has concerns." The sentence is understandable. Anyone who simply blindly copies customer data into ChatGPT's free interface is actually playing Russian roulette with the GDPR. But: the equation "AI = data protection risk" is wrong.

Tesseract vs. PaddleOCR: Strengthening the database for chatbots

Retrieval Augmented Generation (RAG) chatbots are revolutionizing the way we access internal company data. They combine the generative capabilities of Large Language Models (LLMs) with precise information retrieval from a dedicated knowledge database.

AI saves Christmas? How the invisible system could help Santa achieve maximum 'Happiness Performance'.

The reindeer are ready, the sleigh is loaded, and Santa Claus is checking his watch: Only 24 hours to fulfill the wishes of every child in the world. A logistical masterpiece that balances on the edge of the possible every year. But what happens when the magic of the North Pole meets Deep Learning? We take a deep dive into the invisible engine behind the world's most famous logistics manager and show how Artificial Intelligence not only optimizes routes but also redefines the 'Happiness Performance' of the holidays.

RAG is good. GraphRAG ist better. Why vectors are sometimes blind.

Does your AI really understand the connections? We need to talk about the "memory" of our LLM applications. Currently, almost all of them rely on standard RAG with vector databases. This is brilliant for finding similar text modules. But it has a fatal weakness: it is "flat". Vector databases find needles in haystacks. But they don't understand how the haystack is structured.

AI images and copyright: who owns the art from the machine?

Generative artificial intelligence has taken the creative world by storm. But while tools such as Midjourney or DALL-E create impressive works in seconds, the legal system is lagging behind. Two fundamental questions arise: who owns the generated images and is the AI even allowed to train with the works of human artists?

With AI into the premier class of corporate websites.

From digital dust catcher to growth engine. It's time for an uncomfortable confession. Let's be honest: when was the last time you visited a company website and were really impressed?

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.