Skip to content

Data and AI

An automation server for realistic training data

A Node.js server simulating professional correspondence between multiple personas, scheduled over time and delivered by real email, to produce believable training datasets.

Client
Applied research project
Year
2024
Abstract illustration of a generator producing a grid of synthetic data
Multi-persona
simulated people with distinct voices
Scheduled
continuous generation without intervention
Node.js
no heavy infrastructure dependencies

The problem

Training and evaluating systems that work over professional correspondence requires professional correspondence. And a company’s real data cannot be used for that: it contains personal and business information that must not leave its environment.

What we built

An automation server that generates and sends simulated correspondence between several professional personas, each with its own role, tone and schedule, distributed over time so the result resembles a real inbox rather than a synthetic dump.

  • An Express API for defining scenarios, participants and conversation scripts.
  • Scheduling with node-cron, so messages spread across days and working hours instead of arriving all at once.
  • Real SMTP delivery with nodemailer, exercising the whole receiving chain rather than only the content.
  • A reproducible Docker environment so the generator can be brought up on any machine.

Next step

Half an hour well spent

Walk us through the problem on a short call. You leave with a first read on how we'd approach it and what it would involve — no commitment, no sales deck.