Custom GPT

Across Industries

For companies like BMW, Allianz, and Sennebogen, Motius has developed an approach to building AI assistants and agents that can work both on-premise and in the cloud, with data ranging from classified internal documents to public data ingested through web search.

A consolidated data layer ensures governance and reusability, while a flexible platform enables the development of use case-specific AI agents, including third-party solutions. Users can interact via chat app, custom web interface, or IT system integrations.

The platform approach provides enterprise grade scalability and security while still providing flexibility, in concrete technical implementation. The solution will empower Keller & Kalmbach with AI-driven insights and automation to enhance workflows and decision-making.

Demo

Gandalv is an internal product we built to showcase our approach, and help us during the requirements engineering phase of our projects.

Key Features

While the demo shows a specific use case, the platform is designed to be flexible and extensible:

Approach

Knowledge Structuring and Extraction

A team from Motius first analyzes existing content and develops a strategy for providing a structured knowledge base. The strategy may include the following technologies and approaches:

Rapid Prototyping and Integration

Once the data sources have been reviewed and connected, a prototype is developed usually using on-premise LLMs and open source components.

Employees at Keller & Kalmbach can start working with the system and provide feedback after just 1-2 sprints (each 2 weeks).

Data Enrichment and Automation

After the first tests, we add more data sources and build agents that can automate tasks and provide insights for different business processes, such as:

• Accessing knowledge across many systems and regions, asking follow-up questions, and getting relevant documents as sources in the response
• Automating repetitive tasks, such as generating reports, summarizing documents, or extracting key information
• Tool calling with MCP, which allows the AI assistant to act on behalf of the user, but using their credentials and permissions

Architecture

graph TB
    subgraph SS ["Source Systems"]
        ERP[(ERP)]
        CRM[(CRM)]
        EXT[(External Data)]
        FILES[Files]
        OTHER[(...)]
    end

    subgraph DP ["Data Platform"]
        subgraph STORAGE ["Storage"]
            RAW[(Raw)]
            CLEANSED[(Cleansed)]
            ENRICHED[(Enriched)]
            CURATED[(Curated)]
        end

        subgraph SPECIALIZED ["Specialized Storage"]
            VECTOR[(Vector DB)]
            GRAPH[(Graph DB)]
            DWH[(Data Warehouse)]
        end

        subgraph COMPUTE ["Compute"]
            INGEST[Ingest]
            PARSE[Parse/Chunk/Filter]
            EMBED[LLM Model Embedding]
        end
    end

    subgraph AL ["Application Layer"]
        subgraph CUSTOMER ["Keller & Kalmbach GPT Service"]
            SIMILARITY[Similarity Matching]
            LLM_GEN[LLM Model Generative]
            LLM_EMB[LLM Model Embedding]
            INFO_QUERY[Information Querying]
            CHAT_HIST[(Chat History)]
        end

        DASHBOARDS[Reporting Dashboards]
        ERP_APP[ERP]
        CRM_APP[CRM]
        CUSTOMER_APP[Keller & Kalmbach GPT App]
    end

    %% Data Flow
    ERP --> RAW
    CRM --> RAW
    OTHER --> RAW
    EXT --> RAW
    FILES --> RAW

    RAW --> CLEANSED
    CLEANSED --> ENRICHED
    ENRICHED --> CURATED

    CURATED --> VECTOR
    CURATED --> GRAPH
    CURATED --> DWH

    %% Compute Flow
    RAW --> INGEST
    INGEST --> PARSE
    PARSE --> EMBED
    EMBED --> VECTOR

    %% Application Layer Connections
    VECTOR --> SIMILARITY
    SIMILARITY --> LLM_EMB
    LLM_EMB --> INFO_QUERY
    INFO_QUERY --> CHAT_HIST
    LLM_GEN --> INFO_QUERY

    DWH --> DASHBOARDS
    DASHBOARDS --> CUSTOMER
    ERP_APP --> CUSTOMER
    CRM_APP --> CUSTOMER
    CUSTOMER_APP --> CUSTOMER

    %% Styling
    classDef sourceSystem stroke:#0288d1,stroke-width: 2px
    classDef dataStorage stroke:#7b1fa2,stroke-width: 2px
    classDef compute stroke:#ef6c00,stroke-width: 2px
    classDef application stroke:#388e3c,stroke-width: 2px
    classDef specialized stroke:#c2185b,stroke-width: 2px

    class ERP,CRM,OTHER,EXT,FILES sourceSystem
    class RAW,CLEANSED,ENRICHED,CURATED dataStorage
    class INGEST,PARSE,EMBED compute
    class SIMILARITY,LLM_GEN,LLM_EMB,INFO_QUERY,DASHBOARDS,ERP_APP,CRM_APP,CUSTOMER_APP application
    class VECTOR,GRAPH,DWH,CHAT_HIST specialized

Application at Keller & Kalmbach

With this approach, Keller & Kalmbach can start building AI assistants for specific use cases within weeks, while building on an architecture that allows for future growth:

• Cloud and AI-agnostic architecture means Keller & Kalmbach can choose the best providers with no lock-in
• Immediate benefits of AI assistants for specific use cases, such as customer support, internal knowledge management, and process automation
• Go further with AI agents that can automate tasks, and even let your engineering teams build their own agents