System Architecture

Conceptual pipeline and module separation.

EVIDENCE

Pipeline

UMKM→Product Concept Input→Configuration Generator→Constraint Engine→Customer Twin→Virtual Screening→Human Validation→Calibration Engine→Sustainability Screening→MCDM Decision Engine→Sensitivity Analysis→Priority Concepts→Physical Prototype / Market Test

Data Stores

Product Data

Configuration matrix, base product attributes

Persona Data

Illustrative Customer Twin segments

Evidence Library

Literature and official statistics informing the prototype

Human Evaluation Data

CSV-imported or manually submitted respondent evaluations

Calibration Results

Computed only when sufficient human data exists

Module Separation

Simulation, calibration, decision, and UI stay independent

features/configurationsGenerates the full cartesian product of configuration options
features/constraintsFilters infeasible configurations before Customer Twin evaluation
features/customer-twinCustomerTwinEngine.evaluate() — deterministic mock, swappable for a real LLM
features/sustainabilityPrototype Sustainability Index — attribute-based, not LCA
features/human-validationIndependent respondent evaluations, CSV import, and summaries
features/calibrationCross-configuration Spearman ρ, MAE, decision agreement; gated on sample size
features/decisionWeighted Sum Model, Decision Trace, sensitivity, and what-if simulation
core/storeComposition root: persistence adapter + feature use-case orchestration

Before vs After

Potentially reduces unnecessary early-stage physical testing

Before

  1. 1Many concepts
  2. 2Manual selection
  3. 3Physical prototype
  4. 4Customer test

TobaMarketTwin

  1. 1108 configurations
  2. 2Constraint filtering
  3. 3Virtual screening
  4. 4Human validation
  5. 5Calibration
  6. 6Priority concepts
  7. 7Physical prototype / market test

This is a directional claim about workflow shape, not a measured reduction percentage — no empirical evidence yet supports a specific number.