Simulate how
groups deliberate
before making
high-stakes decisions.
Sapiens Q builds pre-decision simulation labs for committees, teams, and populations.
With records and real-world data, modeling how decision-makers deliberate, how teams respond, and how populations may react.
Use them to pre-test policies, surveys, and difficult conversations before they move into the real world.
Three Labs, One Method
Build agents, model decision-makers, simulate populations.
Build Synthetic Agents
Refill
Rehearse sensitive conversations before they happen.
Refill is for managers and people leaders. Build AI counterparts such as a direct report, a teammate, or a stakeholder, and practice difficult conversations in a private, repeatable setting.
It fits performance feedback, one-on-ones, and reviews. Each session can keep context through memory you control, so you can review your delivery and improve over time.
- Performance 1:1s
- Manager L&D
Build Synthetic Agents
Refill
Rehearse sensitive conversations before they happen.
Refill is for managers and people leaders. Build AI counterparts such as a direct report, a teammate, or a stakeholder, and practice difficult conversations in a private, repeatable setting.
It fits performance feedback, one-on-ones, and reviews. Each session can keep context through memory you control, so you can review your delivery and improve over time.
- Performance 1:1s
- Manager L&D
The S.G.E.
Synthetic Generative Environment
We build Synthetic Generative Environments (S.G.E.). These are simulation spaces where agents, rules, memory, and context interact over time.
Each agent is grounded in evidence and role constraints. A cognitive architecture supports memory, reflection, and planning. Its behavior can then be tested against observed outcomes.
Patterns that emerge over time
Agents follow no fixed script, so their interactions can produce new patterns of behavior over the course of a run.
Feedback across each step
Agent behavior and the state of the environment update each other over time, so later steps build on what came before.
The Cognitive Engine
How the Simulation Works
The same engine runs under every lab, and every run can be replayed and tested against ground truth where it exists.
Each agent writes what it observes into a long-term memory stream, then retrieves from it by weighing recency, importance, and relevance together. What it recalls stays tied to a source, so a statement can be traced back to the evidence behind it, and context holds steady across a long session.
Memory that carries forward
Agents keep a long-term memory and draw on the moments that matter when they act.
Each agent writes what it observes into a long-term memory stream, then retrieves from it by weighing recency, importance, and relevance together. What it recalls stays tied to a source, so a statement can be traced back to the evidence behind it, and context holds steady across a long session.
Each agent is built from evidence about whoever it represents and carries its own trait profile. That source might be a figure drawn from the record, a counterpart you define for a rehearsal, or a respondent sampled from population data. As memory accumulates during a run, a reflection step lets that outlook update, so the agent reasons in character and adapts as events unfold. Provenance and non-impersonation guardrails keep each persona tied to its sources and scoped to scenario testing.
Personas grounded in evidence
Each agent is grounded in evidence about whoever it stands in for, and its outlook updates as memory builds.
Each agent is built from evidence about whoever it represents and carries its own trait profile. That source might be a figure drawn from the record, a counterpart you define for a rehearsal, or a respondent sampled from population data. As memory accumulates during a run, a reflection step lets that outlook update, so the agent reasons in character and adapts as events unfold. Provenance and non-impersonation guardrails keep each persona tied to its sources and scoped to scenario testing.
The same engine drives a one-on-one rehearsal, a committee that debates and votes, and a population that answers in parallel. Every run is logged as a replayable trace, and where a real outcome or benchmark exists, that run is scored against it. A decision or an observed distribution becomes the yardstick, so the model can be checked and improved.
Interaction you can replay
Agents interact at any scale, from one counterpart to a whole population, and every run can be replayed.
The same engine drives a one-on-one rehearsal, a committee that debates and votes, and a population that answers in parallel. Every run is logged as a replayable trace, and where a real outcome or benchmark exists, that run is scored against it. A decision or an observed distribution becomes the yardstick, so the model can be checked and improved.