AI services

We build AI into the work your business repeats: agents and language-model features for your product or your team, data extraction from documents, scheduled automations, and AI labelling of collected data. You describe the task; we reply with what we would build, how we would test it and a fixed price; then we build it and hand it over with the test results.

Describe the task

What we take on

Four kinds of work.

AI agents and language-model features

An agent or a language-model feature for your product or your team. It reads, drafts, sorts or answers, returns structured output, and is checked against your rules before anything it produces is used.

Document data extraction

Fields read from PDF and Word documents of one kind into a spreadsheet or your system's import format. Every value is checked against the text it came from, and values that cannot be confirmed are marked for review.

Workflow automation

A recurring task run on a schedule: data pulled from the services you use, checked, and passed on or reported. Every run is logged, and a step that fails raises a notice instead of failing silently.

AI enrichment of collected data

Public web data with fields added by a language model: a category, a score against your criteria or a short summary for each row, each backed by a quoted line from the source.

How it goes

Three steps.

  1. 1

    Describe the task

    Email [email protected] with the task, a few real examples and where the result should go.

  2. 2

    We reply with a scope

    What we would build, how we would test it on your examples, the running cost per item, and a fixed price.

  3. 3

    We deliver

    The working build, its test results on your examples, and notes on how to run and change it.

Shown working

From a task to a tested build.

Scope, build, test, cost, hand over.

from task to hand-overfive steps

  1. Scopethe task and your examples
  2. Buildprompts, code and checks
  3. Testscored on your examples
  4. Costmeasured per item
  5. Hand overcode, notes and results

What we have built with these methods

These are our own builds, not client work.

  • Three data tools on Apify, written, tested and published by AI agents, with every change recorded in version control.
  • A scheduled pipeline that measures site traffic, plans one task, drafts with a language model and logs every run, with a time limit on each stage and nothing published until a gate is opened.
  • A live speech pipeline: speech recognition, language-model translation and a synthetic voice, with a spending limit on each session.
  • Language-model screening of collected documents against a fixed list of requirements, each verdict backed by a quote that is checked automatically against the source text.

Your data and your systems

This work runs on the documents, data and access you give us, and we use them only for your job. Before we start, we tell you which AI provider will process your data. There is no price list: each job gets a fixed price in the scope reply.