← Back to work

AI · Case study · Private

ParticleX

A synthetic experimentation engine - run A/B tests on AI-generated user personas before risking real traffic.

LovableLLM-based simulationPersona modeling
internal
ParticleX preview

Context

The setup before I touched it.

Real A/B tests are expensive: they need traffic, engineering effort, and weeks of waiting. Most copy and UX decisions never get tested at all because the cost is too high relative to the stakes.

Problem

  • 01Teams ship copy and design changes on instinct, not evidence.
  • 02Real experiments can't cover the long tail of small decisions.
  • 03Segment-level insight is buried until the experiment ends.

Approach

  • 01Generate a population of synthetic users with realistic demographics, intent, and decision biases.
  • 02Run any variant (copy, page, offer) past that population and aggregate predicted behavior.
  • 03Surface uplift estimates and segment-level reactions, not just a single winner.
  • 04Treat outputs as a pre-flight check - point teams at the variants worth testing for real.

Outcome

  • 01Faster cycles between idea and decision.
  • 02Lower-stakes decisions get evidence instead of opinion.
  • 03Teams enter real experiments with a hypothesis, not a coin flip.

Role

Concept, design, and build.

Status

Live - internal

Next

Browse more work →