About PolicyForge

Decision support, not autonomous policymaking.

PolicyForge is a sandbox for exploring how a policy change might affect a Chennai-calibrated synthetic population. It does not decide policy or predict specific people.

OBSERVED DATA — sourced contextSYNTHETIC AGENTS — generatedSIMULATION RESULTS — model output

AI agents are simulations, not humans.

Their behavior follows explicit model assumptions, not real cognition or real lived experience.

Synthetic agent data is labelled.

Chennai population totals and city-service context are observed; individual agents and their behavioural variables are generated by the model.

Uncertainty matters.

Seeds, parameters and model structure can change outcomes. Results should be compared across scenarios, not treated as certain predictions.

Evidence boundaries matter.

Observed data, simulation results and model-inferred insights remain distinct throughout the product.

Sensitive traits do not determine behaviour.

Population characteristics guide aggregate calibration, never fixed individual outcomes.

Learning & calibration

Make the model auditable.

PolicyForge separates what is observed from what is generated and what the simulation produces.

1Establish contextUse the Chennai evidence layer to inspect the aggregate data available for calibration and scenario design.

2State assumptionsChoose the synthetic population preset, policy parameters, simulation horizon and random seed.

3Compare outcomesRead model outputs alongside uncertainty ranges and rerun with different seeds or scenarios.

Important limitationCalibration anchors the context; it is not a real-world forecast. Behaviour and policy effects remain explicit synthetic-model assumptions.

Project team

Developed by

Mahadevan RajagopalanAsvath MSanjit SKrish MuralidharanSai retheka