Skip to content
A2H-Consulting
Popular searches
    Start a project
    Menu
    ServicesProjectsProcessAboutDeutsche Version

    AI & Agents

    AI applications that hold up in daily use — on your data.

    We build RAG assistants, document screening and extraction — on Azure or Databricks, with answer-quality evaluation and a clear sign-off process.

    Typical situations

    • A chatbot prototype is convincing, but security, cost and operations are unresolved.
    • Business teams still check contracts, invoices or certificates by hand.
    • Knowledge is scattered across manuals, SharePoint, file shares and business systems.
    • Several teams are building similar AI solutions in parallel.
    • Answers from an AI assistant cannot be traced — sources, hallucinations and quality are not measured.
    • Data protection and the works council require a concept before AI may go into production.
    • Business teams spend more time searching than deciding.
    • First GenAI solutions are running, but cost per request and value are unknown.

    What we deliver

    • RAG assistants with source citations, access controls and a feedback loop
    • Document screening: review rules, deviation detection, hand-off to business systems
    • Extraction and classification with Document Intelligence and language models
    • Agents and assistants that prepare process steps — people make the decision
    • Evaluation (answer quality, hallucinations), monitoring, cost control
    • LLMOps: versioning of prompts and models, automated tests and CI/CD
    • Reusable building blocks so the next use case goes faster
    • Security and data protection concept, including prompt-injection protection and an approval process

    Technologies

    • Azure OpenAI
    • Microsoft Foundry
    • Azure AI Search
    • Document Intelligence
    • Databricks Mosaic AI
    • Vector Search
    • Model Serving
    • MLflow
    Outcome

    AI applications that business teams use and IT can take responsibility for.

    AI in practice

    Use cases we take into production.

    On Azure or Databricks, we start with a specific use case and build the platform so the next one goes faster.

    • Knowledge assistant (RAG)

      Ask questions of manuals, policies and project knowledge — with source citations and access controls.

    • Document screening

      Check contracts, invoices, certificates or applications automatically and flag deviations.

    • Extraction & classification

      Capture data from PDFs, scans and emails in structured form and pass it on to SAP or business systems.

    • AI for master data quality

      Detect duplicates and inconsistencies before they reach SAP.

    • AI agents & Copilot

      Agents that prepare requests, checks or orders across systems — and a Copilot that knows your knowledge base.

    • Forecasting & Power BI

      Predict demand, quality or costs and make it visible to everyone in Power BI.

    Every use case comes with answer-quality evaluation, monitoring and a clear sign-off — not just a demo.

    How we take an AI use case into production

    1. 01approx. 1 week
      Use case checkAssess goals, value, data sources and risks together with the business team.
      You get
      • Use case brief with value and success criteria
      • Assessment of data, feasibility and data protection
      • Effort and cost estimate, clear go/no-go recommendation
    2. 02approx. 2–4 weeks
      Prototype on real dataA working prototype in your environment, measured against an agreed test set.
      You get
      • Working prototype to try out
      • Quality measurement: hit rate, answer quality, failure cases
      • Architecture proposal and cost per request
    3. 03approx. 4–8 weeks
      Production rolloutSecurity, permissions, monitoring and sign-off — rolled out as code across DEV, UAT and PROD.
      You get
      • Production application in your Azure or Databricks environment
      • Infrastructure and deployment as code (Terraform, CI/CD)
      • Monitoring, cost control and sign-off record
    4. 04ongoing
      Handover & operationsKnowledge transfer to your team and support through the first weeks of operation.
      You get
      • Documentation, runbooks and training
      • Quality dashboard for ongoing operations
      • Prioritized list of next use cases
    Overall outcomeA production AI application that runs in your environment and belongs to you — with code, documentation, measurable quality and reusable building blocks so the next use case goes faster.

    More services