Synthetic Signals
← Platform
Personas & audiences

Test on the people you actually serve.

Every citizen is a persona you can talk to — grounded in real Census data, with a personality and memory. Tune every trait, assemble an audience for the exact job you're testing, and report on the outcomes that matter to you. Not the 20 personas you'd write by hand — thousands, and you choose who.

AudienceApplying for a loan
PN
Priya Nair58 · first-time borrower
MM
Marcus Meyer34 · debt consolidation
GW
Grace Wong71 · ESL · fixed income
420 personas · built for one job-to-be-done
Thousandsdistinct personas in one audience
~2.7%off from real Census demographics
100%synthetic — no real people, no personal data
One jobpurpose-built audiences for a job-to-be-done
Personas

Every citizen is a persona — not a script

A persona here isn't a paragraph you wrote. It's a synthetic person with Census-grounded demographics, Big Five (OCEAN) personality, a daily life and memory across sessions. Talk to one, or run your agent against thousands at once — the skeptic, the rambler, the anxious caller, all in character.

  • Census demographics, OCEAN personality and memory per persona
  • Talk to one, or the whole population in parallel
  • Hard cases show up in character — without hand-authoring each one
HS
Hana Singh32 · Female · Healthcare
Income$100–150k
HouseholdMarried · 2
EducationGraduate
CommuteBus · 28 min
Big Five (OCEAN) personality
O
C
E
A
N
Custom audiences

Assemble an audience for a specific job

An audience here isn't just a demographic skew — it's built for a job-to-be-done. Loan applicants, parking-permit renewals, first-time buyers: describe the scenario and the audience assembles from citizens whose lives actually fit it. Need a market shape instead? Tilt age, income or language and the population re-weights while staying Census-honest.

  • Purpose-built for a job-to-be-done, not just a skew
  • Or tilt any axis — age, income, language, household
  • Save an audience and reuse it across runs
AudienceApplying for a loan
PN
Priya Nair58 · first-time borrower
MM
Marcus Meyer34 · debt consolidation
GW
Grace Wong71 · ESL · fixed income
420 personas · built for one job-to-be-done
Custom personas

Define the personas you need

Add attributes and behaviors we don't ship out of the box — a chronic condition, low digital literacy, a distrust of AI, a tendency to skip follow-ups. Build the difficult, high-stakes, easy-to-miss people your agent has to get right, layered on top of the population you already have.

  • Add custom attributes on top of the Census base
  • Add behaviors that shape how a persona acts
  • Build the edge cases you'd never think to script
Attributes
+ chronic condition+ insurance plan+ digital literacy
Behaviors
distrusts AIcode-switchesskips follow-ups
Grounded in
your region · ACS+ your aggregate stats
Match your users

Match your real user base

Share the shape of your audience — segment sizes and key distributions, aggregate stats, never individual records — and we synthesize a population that mirrors it. No PII ever changes hands, and the result stays fully synthetic, just like everything else.

  • Match your users from aggregate stats — never records
  • No individual data, no PII — fully synthetic output
  • Reproducible: the same audience comes back every run
Synthetic populationvs. ACS marginals
Age 18–34
31%
Age 35–64
49%
Age 65+
20%
< $50k
28%
$50–150k
44%
$150k+
28%
Fit error 2.68% — matches the real city
Custom reporting

Measure what matters to you

The triad finishes with reporting: your rubric, your cohorts, your dashboards. Score with the built-in model-graded rubric or bring your own evals, break results down by the segments you care about, and export everything — transcripts over MCP or REST, runs as OpenTelemetry traces — into the stack your team already works in.

  • Your rubric or BYO scoring — never a black box
  • Cohort breakdowns for tests and interviews alike
  • Exported to your stack — MCP, REST, OpenTelemetry
Task success
95
Consistency
92
Satisfaction
95
Safety
88
or pull transcripts over MCP into your own evals
How it works

From a base city to the exact audience you need.

Start from a Census-grounded city and shape it down to the people who matter for your agent.

1
Choose a base

Pick a city and population size — a Census-grounded starting point.

2
Shape the audience

Describe the job-to-be-done, or skew age, income and language to match your market.

3
Add custom personas

Layer on the attributes and behaviors your use case needs.

4
Run and break it down

Set your agent loose and read results by persona and cohort.

Grounded in US Census data (ACS) & OpenStreetMap. Synthetic throughout — no real individuals, no personal data.

Questions

Personas, audiences, and what you control.

Are these real people?

No. Every persona is fully synthetic, generated from public, aggregate Census data. There are no real individuals, no personal data, and nothing that maps back to a real person.

What's the difference between a persona and an audience?

A persona is one synthetic person — demographics, personality and memory. An audience is a segment of many personas, assembled for a purpose: a job-to-be-done like applying for a loan, or a market shape like a senior-heavy city. You can talk to a single persona or run your agent against a whole audience at once.

What's a purpose-built audience?

An audience assembled around a scenario rather than a demographic slider: describe the job-to-be-done — disputing a fee, renewing a permit, comparing plans — and the audience draws citizens whose circumstances actually fit that job, while staying demographically honest.

How is this different from the QA personas we write today?

Hand-written personas only cover the handful of people you already thought of. These are thousands, demographically grounded, and shaped to your market — so you also find the cohorts you'd never have scripted.

Can I define my own persona or match my real users?

Yes. Add custom attributes and behaviors on top of the Census base yourself. To match your real users, share aggregate stats — segment sizes and key distributions — and we synthesize a population that mirrors them. We work from aggregates, never individual records, so no PII changes hands and the output stays fully synthetic.

Is a custom audience reproducible?

Yes. The population and setup are seed-deterministic, so the same audience comes back every run. See Reproducibility for what is and isn't deterministic.

Test on the people who actually matter.

Shape your audience, add the personas you need, and set your agent loose.