AI Solutions

AI solutions that fit your processes

We build AI assistants, automations and custom models that fit securely into your systems and stay traceable and under your control.

GAD is certified to ISO/IEC 27001:2022 (No. 26052601). More on security & compliance

·WHAT WE BUILD

Six building blocks that make up your AI project

Every AI project is made of a few proven building blocks. We combine them to fit your processes, systems and data. Designed for GDPR-compliant use and, if you wish, fully on your premises.

  • 01AI assistant on your documents

    Documents › Search › Answer with source

    Answers questions from your manuals, files and policies and cites the source. It is based on RAG: the system first finds the matching passages and only then writes the answer.

    Technology: RAG, permission concept

  • 02Workflow automation

    Trigger › Rules › Action

    Recurring processes run by themselves: transferring data, collecting approvals, notifying people. Among other tools we use n8n, an open-source tool that connects programs and runs on your own servers.

    Technology: n8n, interfaces (APIs)

  • 03Document processing

    Inbox › Extract › Business system

    Invoices, applications and forms: the AI reads the relevant fields, checks them and passes them to your business system. Unclear cases go to a person.

    Technology: Text recognition (OCR), checking rules

  • 04AI agents

    Task › Plan › Done

    An agent does not just answer, it carries out steps: creating a ticket, booking an appointment, looking up data. It may only do what you allow, and every step is logged.

    Technology: Permissions per tool, audit log

  • 05Connection to your systems

    Email · ERP/CRM · DMS · Teams

    We connect the AI to email, ERP and CRM (resource planning and customer management), DMS (document storage) and Microsoft Teams. Your team keeps working in the tools it knows.

    Technology: Standard interfaces

  • 06Operation on your premises or in Germany

    On your premises or in Germany

    The system runs on-premises, meaning on servers in your building, or in a data center in Germany. You decide where your data is stored.

    Technology: Open models, encryption

Examples from your industry

Choose your industry. You will see what we typically build there and which building blocks are used.

Contract pages with passages highlighted in green next to a laptopLaw firms

  1. Sort incoming mail and record deadlines

    Briefs and emails are assigned to the right case file, deadlines are proposed as calendar entries.

    Building blocks: Documents, workflow

  2. Check contracts against your templates

    Deviations from your standard clauses are marked with their location. You review exactly those passages.

    Building blocks: Assistant

  3. Research your own knowledge base

    Ask questions about the firm's briefs, opinions and templates and get answers with sources.

    Building blocks: Assistant, on-premises

Claim documents with a photo of a damaged car next to a laptopInsurers

  1. Pre-fill claim notifications

    Forms, photos and emails are read and the case is created in the policy system.

    Building blocks: Documents, connection

  2. Pre-sort customer requests

    Emails are classified by topic, forwarded to the right team and given a draft reply.

    Building blocks: Workflow, agent

  3. Query policy conditions

    Staff ask, the assistant answers from the valid policy terms with the exact reference.

    Building blocks: Assistant

Bright file archive with shelves full of folder boxesPublic sector

  1. Check applications for completeness

    Missing documents are detected, the follow-up request is ready as a draft. The case officer decides on the application.

    Building blocks: Documents, workflow

  2. Answers from files and regulations

    Every answer with its source. Each person only sees what their role allows.

    Building blocks: Assistant

  3. Run everything in-house

    Models and data stay on your servers, with no connection to external AI services.

    Building blocks: On-premises

Machine spindle with sensor and green status LED in a production hallManufacturing

  1. Delivery notes and invoices into the ERP

    Documents are read, matched with the purchase order and prepared for booking.

    Building blocks: Documents, connection

  2. Maintenance knowledge on demand

    Manuals and fault reports as an assistant for maintenance, also on a tablet on the shop floor.

    Building blocks: Assistant

  3. Flag deviations in machine data

    A model learns normal operation and opens a ticket or posts in Teams when something deviates.

    Building blocks: Agent, workflow

AI-generated image: Employee seen from behind in front of a screen with a trend lineAI contentBanks

  1. Read loan documents

    Values from payslips and balance sheets are captured, the checklist is pre-filled. Your team makes the credit decision.

    Building blocks: Documents

  2. Review unusual transactions first

    The AI finds transactions that deviate from the usual pattern (anomaly detection). Your team decides.

    Building blocks: Agent, connection

  3. Query regulations

    Internal policies, MaRisk and DORA: answers with references instead of long searches.

    Building blocks: Assistant, on-premises

AI-generated image: Team seen from behind at a whiteboard with a sketched processAI contentOther industries

  1. Prepare quotes from requests

    The request email is read, line items are recognized and a draft quote is created in the CRM.

    Building blocks: Documents, connection

  2. Answer internal questions

    IT and HR policies as an assistant directly in Microsoft Teams.

    Building blocks: Assistant, connection

  3. Stop retyping data

    Recurring transfers between programs run automatically.

    Building blocks: Workflow

Your process is not listed? We will look at it and tell you honestly whether AI helps there.

Describe your process in a few steps

We will get back to you with questions and a proposal for the building blocks that fit.

Discuss your use case →Contact us directly
·AI DIAL

How much should AI take over in your company?

AI does not have to do all or nothing. Set for each department how independently it should work: no AI, suggest (the AI prepares, your team decides), approve (a person approves every step) or autonomous (your team checks samples). The preset is a careful starting point.

Together we check which level fits, including GDPR and the EU AI Act (the European rules for AI systems).
Request a call
·AI TEST AUTOMATION

Test AI before it creates risk

Before an AI system goes live, we test it automatically against a fixed set of test cases, repeatably and later in operation too. You see where it answers wrongly, can be manipulated or gets worse over time.

Your AI system MODEL, LLM, DATA Correct answers QUALITY Manipulation SECURITY Bias FAIRNESS Performance over time OPERATION Audit report EVIDENCE Automated test run repeatable, before and after go-live Release with evidence Your AI system MODEL, LLM, DATA Correct answers QUALITY Manipulation SECURITY Bias FAIRNESS Performance over time OPERATION Audit report EVIDENCE Automated test run Release with evidence

ML models

Classic AI for forecasts and classification. We measure accuracy, stability and drift, meaning whether performance declines in operation.

Result: You notice early when the model gets worse.

Language models (LLMs)

We test for made-up answers (hallucinations), steering through manipulated inputs (prompt injection) and data leakage.

Result: The assistant only goes live once these tests are passed.

Data

We check training and production data for gaps, errors and one-sided distribution (bias).

Result: The model does not learn from faulty data.

Request AI Test Automation →
·CUSTOM MODELS

A model of your own, trained on your data

A general chatbot does not know your machines, cases or forms. For tasks like these we train a specialized model with your own examples. A model is a program that learns patterns from examples and recognizes them in new cases.

Your data stays on your premises or in a data center in Germany.

Bright server room with an open server rack and green status LEDs

What for, for example

  • Failure early warning

    The model learns from sensor data what a normal run looks like and reports deviations before a machine fails.

  • Unusual claims

    It flags claims that resemble earlier fraud cases. A person makes the decision.

  • Route incoming mail

    It assigns incoming letters and emails to the right case and team.

  1. 01

    Review data

    We check which examples exist, such as sensor readings, files or emails. The data stays in your environment

  2. 02

    Training

    The model learns from your examples, on your servers or in a data center in Germany

  3. 03

    Test

    Before use, we check accuracy, errors and bias, as with AI Test Automation above

  4. 04

    Deployment

    The model works inside your process, for example as an alert in the control room or a suggestion in the ticket system

Start: your existing dataResult: a model for exactly your task
Request a custom model →

Questions about AI solutions? Answers

What is RAG?

RAG stands for retrieval-augmented generation. The system first finds the matching passages in your documents and only then writes the answer. This is why the assistant names the source for every answer.

Where is our data stored?

You decide where your data is stored. The system runs on-premises, meaning on servers in your building, or in a data center in Germany. When run in-house, models and data stay on your servers, with no connection to external AI services.

How does an AI assistant differ from an AI agent?

An assistant answers questions from your documents. An agent completes steps itself, such as creating a ticket, entering an appointment or looking up data. It may only do what you approve, and every step is logged.

Does the AI decide on its own?

No. The AI prepares, your team decides. Unclear cases go to a human. A case officer still decides on an application, your credit team on a loan.

Which programs can the AI be connected to?

We connect the AI to email, ERP and CRM, to document management (DMS) and to Microsoft Teams. Your team keeps working in the programs it knows.