---
title: "Architecture makes the difference."
description: "Connecting an LLM is the easy part. As chatbot developers, we build the architecture, RAG pipeline and interfaces so your bot works with real data."
url: https://www.lautmaler.de/en/services/development
lang: "en"
alternate_de: "https://www.lautmaler.de/services/entwicklung"
---
[Services](https://www.lautmaler.de/en/services) Development

04 / Development

# Architecture makes the difference.

Connecting an LLM is the easy part. As chatbot developers, we build the architecture, RAG pipeline and interfaces so your bot works with real data.

[Discuss a project](https://www.lautmaler.de/en/contact)[See all services](https://www.lautmaler.de/en/services)

![Illustration: Entwickler am Schreibtisch vor einer Systemarchitektur mit verbundenen Modulen](https://www.lautmaler.de/uploads/Dizw8C9cxSI3ZDVRoU-x2-svc-development-scene.png)

Case study

## In practice

[Discuss a project](https://www.lautmaler.de/en/contact)[View the YOUNA project](https://www.lautmaler.de/en/projects/youna)

YOUNA

/ First aid against racism

Rigid rules became a bot you can talk to.

YOUNA launched in 2022 with a rule-based bot. For people who had just experienced racism, it was too rigid. Step by step, we rebuilt it into a controlled language model that only draws on expert-reviewed knowledge and makes nothing up. It has been live since December 2024. Conversations now last 90% longer, and around twice as many people start one at all.

54%

**don't trust answers from AI.** That's why our bots answer from your real data.

31%

have already received **wrong information from an AI**. That's why we test with real questions before the bot goes live.

74%

of companies have **pulled an AI agent again after launch**.

/ What's included

## What our chat- and voicebot development includes

We build the architecture, integrate LLMs and voice and connect the bot to your CRM and your website. Then we test until it runs under real load too.

### Architecture & bot logic

We choose the architecture by your requirements, your scale and your data protection. Whatever happens to be trending plays no part. We implement the dialogue logic so the bot behaves as planned in the concept.

### NLU & LLM integration

We train the bot to recognize what someone wants and which details belong to it. Technically, that's intents and entities. For LLM integration we use prompt engineering, prompt chaining and a RAG pipeline. That way the answers rest on your data, not on guesses.

### Voice integration (ASR & TTS)

We integrate speech-to-text reliably and fine-tune text-to-speech until the bot's voice sounds natural and pronounces company-specific words correctly.

### Backend & API integration

CRM integration or identity management, we connect your systems via API and integrate the chat- or voicebot into your website.

### Testing & performance

We combine automated tests with manual review and keep optimizing answer quality and performance.

/ FAQ

## Frequently Asked Questions

Here you'll find the questions we're asked most often. Can't find what you're looking for? **Then just chat with our bot!**

[ChatLord](https://www.lautmaler.de/en/contact)

**How do you integrate ChatGPT, Claude or other LLMs?**

For LLM integration we rely on prompt engineering, prompt chaining and RAG pipelines. That way the LLM answers reliably, on brand and based on your real data.

**What is a RAG pipeline, and why do we need one?**

RAG (retrieval-augmented generation) lets an LLM answer from your own data instead of only from its training knowledge. That's the basis for reliable, up-to-date answers.

**Can the bot connect to our existing CRM or other systems?**

Yes. Through API and backend integration we connect systems such as CRM or identity management, so the bot works with real data.

**Can you build voicebots too, not just chatbots?**

Yes, but a voicebot isn't a chatbot with a voice! Spoken conversations need their own design: shorter sentences, no lists or buttons, and instead a plan for pauses, interruptions and cases where the bot mishears something. On the technical side, we integrate ASR and TTS and tune the voice until it sounds natural rather than mechanical.

**Which chatbot architecture do you use?**

There's no standard solution. We choose it by your requirements, scale and data protection instead of following a fixed framework. We build new bots as well as extensions to existing systems.

**How do you ensure quality?**

Through automated and manual testing. That way we catch technical regressions and real usability problems. We optimize performance and answer quality continuously, not just before launch.

## From concept to a bot that actually runs.

![Co-Founder & Managing Director Technology - Implementation, Information Architecture, Backend Integration, API Integration, Data Visualisation, Business Consulting](https://www.lautmaler.de/uploads/2Knki6nQUDF6XrijJFQL3-team-core-1.webp)

Your contact

Norbert Gocht

/ Co-Founder & Managing Director Technology

[Contact](https://www.lautmaler.de/en/contact)
