What are you looking for?

Start here

    Contact us
    Menu

    TECHNICAL APPROACH

    What happens behind the workflow?

    A closer look at where AI, rules, connections and human review fit. Technical detail should help you make a decision and understand the work.

    Explore each service Try the demos

    THE WORKFLOW, END TO END

    AI is one step in the system.

    Each service has its own flow. The shared approach is to bound the input, use AI where useful, check the output and finish with a clearly permitted action.

    1. 01

      Receive the input

      A form, message, text-based PDF or short voice recording.

    2. 02

      Check the conditions

      Required fields, format, size and permitted information.

    3. 03

      Use AI where useful

      Interpret meaning, extract information or prepare a draft.

    4. 04

      Validate with rules

      Check fields, sources and amounts; surface uncertainty.

    5. 05

      Route the output

      Present a result, ask for review or use an agreed connection.

    INSIDE THE SIX SERVICES

    Inputs, logic and boundaries.

    These details describe the website examples and the basic service approach. Tools and action permissions for a business implementation are defined separately.

    01

    Sales & customer follow-up

    Input
    A browser conversation is submitted with the permitted enquiry fields.
    Where AI helps
    The model interprets the message and extracts the details needed for qualification.
    Defined rules
    The server validates required fields and applies deterministic priority and routing rules.
    Output
    A structured result is shown for review and can be downloaded locally.
    Boundary and scope
    The demo has no customer-system write, email sending or booking connection.
    02

    Email management & reply preparation

    Input
    Fictional email text is entered in the browser.
    Where AI helps
    The model proposes a category, summary, missing details and reply text.
    Defined rules
    The server checks the response structure and the permitted review-only scope.
    Output
    The browser displays an editable draft with copy and rejection actions.
    Boundary and scope
    Copying or approving the local text does not create or send an email in an account.
    03

    Document & order processing

    Input
    The deployed service extracts selectable text from the supplied PDF.
    Where AI helps
    The model identifies fields and the source text supporting them.
    Defined rules
    The server validates source evidence, required fields and arithmetic; approval is tied to the reviewed version.
    Output
    A reviewed result can be exported as CSV.
    Boundary and scope
    The demonstrated scope covers text-based samples and a local file export; it has no accounting-system write.
    04

    Company knowledge & text support

    Input
    A question and its selected language enter a public-content-only flow.
    Where AI helps
    The assistant finds relevant approved company information and can use it to compose a conversational answer.
    Defined rules
    The server applies fixed privacy and action safeguards, verifies that cited sources are approved, and returns an unknown or human-review result when no reliable source is available.
    Output
    A conversational answer with its public source when available, or a clear unknown or human-review result.
    Boundary and scope
    Answers stay within approved public company information. The assistant cannot access private customer data or send a referral.
    05

    Price estimates & pro forma invoices

    Input
    A written need and the options selected in the browser.
    Where AI helps
    The model suggests service options; the user decides whether to apply them.
    Defined rules
    The server applies the example price list and agreed calculation rules, counting shared components once.
    Output
    PDF and JSON files are prepared from the same calculation result.
    Boundary and scope
    Educational totals are not a commercial offer; email dispatch and order acceptance are outside the demo.
    06

    Website voice assistant

    Input
    After microphone permission, the server establishes the connection and the browser streams live audio over WebRTC. The permanent provider key stays on the server.
    Where AI helps
    A realtime audio model uses approved company information. Semantic turn detection recognises when the visitor has finished speaking and permits interruption.
    Defined rules
    The server supplies the approved information and assistant instructions. The permanent provider key stays on the server.
    Output
    The browser plays responses as they arrive, with mute and end controls and an optional transcript.
    Boundary and scope
    This browser conversation does not connect a telephone line, transfer a call, create a booking or write to company systems.

    CURRENT TECHNICAL FOUNDATION

    A focused interface. Defined server-side work.

    This website uses Astro and TypeScript. Interactive examples use server functions and OpenAI services where processing is needed. Each service defines checks and output formats suited to its task.

    This does not mean every customer project must use the same tools or that every existing application can be connected directly.

    Before connecting a business tool

    • What do the source and destination applications allow?
    • What information is actually needed?
    • Which access permits reading, recording or sending?
    • How is a successful write confirmed?
    • How are errors, retries and usage limits handled?

    START WITH YOUR WORK

    Describe the task. Let’s make the first step clear.

    Tell us what you need Read the FAQ