shouldi.io
AI-based Decision Simulator
Submit any "Should I...?" question and receive a multi-perspective outcome analysis powered by domain-specific AI advisors and a client-side Monte Carlo simulation engine.
The Challenge
Decision-making typically relies on intuition and incomplete information. We developed shouldi.io to democratize access to analytical decision support — a capability previously available only through expensive consultants or complex analytical tools. The platform needed to handle vastly different decision domains (finance, health, career, relationships) while returning structured, probabilistic results fast enough to feel instant.
Our Approach
The platform uses Google Gemini (gemini-2.0-flash-lite) via Next.js 16 server actions to power eight domain-specific AI advisor personas — covering finance, career, health, relationships, education, real estate, lifestyle, and business — automatically selected based on the user's question. Each advisor generates tailored survey questions and feeds into a client-side Monte Carlo engine (lib/monte-carlo.ts) that runs 1,000 Gaussian-sampled iterations to produce confidence bands and risk scores across best, moderate, and worst-case scenarios. Firebase Auth (email/password + Google OAuth) and Firestore handle authentication and encrypted, user-scoped storage.
Key Features
Monte Carlo Simulation — 1,000 iterations with Gaussian sampling produce confidence bands and risk scores per scenario. Domain-Aware AI Advisors — eight personas auto-detected from input, each generating customised analysis. Multi-Scenario Visualisation — interactive histograms for best, moderate, and risk outcomes. Crisis Detection — self-harm phrase recognition (EN + DE, leetspeak-normalised) redirects to resources before any other processing. Multilingual — full English and German interface via next-intl. PDF Export & Dashboard — shareable results and history of past analyses.
Results & Impact
Over 1,000 users have run analyses with sub-5-second response times despite each query triggering 1,000 Monte Carlo iterations client-side. The platform maintains full user privacy through encrypted, user-scoped Firestore architecture and is deployed on Vercel for consistent global availability.