Scenario interface
Amplify · API Gateway · Lambda
Accepts scenario input, validates, initialises the simulation run, and returns a run_id immediately. All subsequent interaction is async.
→Platform Architecture
A simulation engine built
on institutional reality.
Six interdependent systems — from real-time scenario ingestion to a permanently running calibration loop — forming a digital twin of the US legislative process. Click any component to explore its design.
AWS-native · Bedrock + Step Functions
535-agent population
SRA v2 · 2026
Confidential — investor reference
Interactive architecture — click any component
Click any highlighted component to expand its detail below
Six core components
Simulation orchestrator
Step Functions Express
Owns the simulation clock. Dequeues events, enforces World Model rules, resolves agent scope, fans out parallel turns. Never calls Bedrock.
→World model
Lambda · deterministic · no Bedrock
Eight pure rule functions encoding real legislative procedure — committee referral, vote tallying, Byrd rule, session clock, discharge petition.
→Agent population runtime
Lambda · Bedrock Haiku + Sonnet
Four-step per-agent turn: Haiku retrieves and evaluates context against persona, Sonnet decides and narrates, DynamoDB confirms before events emit.
→Neptune Analytics
Political knowledge graph
Graph of every relationship between actors, donors, committees, bills, and media. Seven node types, seven edge types, continuously refreshed from 14 data sources.
→Continuous learning loop
EventBridge · Lambda · DynamoDB
Permanently running background system. Every real-world vote recalibrates agent behavioral weights. The accuracy advantage compounds with time-in-market.
→The full technical specification — agent JSON schema, event registry, data source registry, and calibration signal functions — is available under confidentiality governance to qualified investors and design partners.