Conversational intake, extraction, vision

AI intake assistant

An intake assistant: a customer describes a leaking coffee machine by text and by voice, each answer fills a structured request field by field, the assistant asks which Thursday is meant, the customer corrects the quantity and confirms the summary, then the request passes its checks and a barcode label is printed.

  1. Multi-turn chat, typed or spoken
  2. Extracts fields, asks about missing details
  3. Each field checked after the customer confirms
  4. Barcode label laid out in millimetres

Language

  1. General chatbot

    Multi-turn intake, confirmed before it is final

    • Next.js
    • React
    • Python
    • FastAPI
    • Pydantic
    • OpenAI API
    • JSON Schema

  2. Structured extraction

    Text or speech into fields

    • Python
    • FastAPI
    • Pydantic
    • OpenAI API
    • JSON Schema

Engineering

  1. AI-assisted development

    Spec-bound generation, delivered in steps

    • OpenAPI
    • DBML
    • Mermaid
    • Python

  2. Output checks

    APIs, models and copy checked item by item

    • Python
    • OpenAPI
    • Redocly
    • DBML

Vision

  1. Images and barcodes

    Line art and label layout

    • Python
    • OpenCV
    • NumPy
    • JsBarcode

  2. Camera capture

    Preview, photos, recording

    • Python
    • FFmpeg
    • PyAV
    • FastAPI

  3. Interactive 3D

    Inspect models in the browser

    • React
    • Three.js
    • WebGL

Models

  1. Modelling prototype

    Baselines and error comparison

    • Python
    • pandas
    • scikit-learn
    • Matplotlib
Ask about this