DecisionLab

Desenhe, simule e audite políticas de decisão antes de colocá-las em produção.

Design, simulate and audit decision policies before putting them into production.

PT-BR · English · Live Demo · Stack · Architecture · Quick Start · Author

Live Demo · Repo · Portfolio · LinkedIn

Lab / demo notice: the public Vercel lab runs a frontend-first simulator (synthetic cases + demo policies). FastAPI remains available locally. This is not an automated production decision engine for individuals.

---

PT-BR

Visão geral

O DecisionLab é um laboratório para construir políticas (regras + threshold), simular custo esperado, explorar sensibilidade, comparar versões e gerar trilha de auditoria + memo executivo.

Problema

Descontos, priorização de leads, revisão de pedidos e acionamento de risco vivem em planilhas e intuição — sem matriz de custo nem audit trail antes do rollout.

Para quem

Analistas de decisão, ops/risk product e data professionals que precisam comparar políticas antes da produção.

Funcionalidades

• Policy builder (regras, threshold, ações)
• Simulador com custo esperado, precision/recall e volume
• Matriz de custo e explorador de threshold
• Painel de fairness (lab) e comparação de versões
• Audit trail + executive decision memo (Markdown)
• 3 políticas demo + casos sintéticos no browser
• API FastAPI opcional (apps/api) com testes

Escopo e limites (honestos)

• Não automatiza decisões reais sobre pessoas
• Demo pública = engine client-side + JSON sintético
• Fairness/cost são suporte à discussão, não compliance certificação
• Sem auth multi-tenant / feature store enterprise

---

English

Overview

DecisionLab is a lab to build policies (rules + threshold), simulate expected cost, explore sensitivity, compare versions and produce an audit trail + executive memo.

Problem

Discounts, lead prioritization, order review and risk triggers often live in spreadsheets and intuition — with no cost matrix or audit trail before rollout.

Who it is for

Decision analysts, ops/risk product folks and data professionals who need to compare policies before production.

Features

• Policy builder (rules, threshold, actions)
• Simulator with expected cost, precision/recall and volume
• Cost matrix and threshold explorer
• Fairness panel (lab) and version comparison
• Audit trail + executive decision memo (Markdown)
• Three demo policies + synthetic cases in the browser
• Optional FastAPI package (apps/api) with tests

Scope and honest limits

• Does not automate real decisions about individuals
• Public demo = client-side engine + synthetic JSON
• Fairness/cost support discussion — not compliance certification
• No multi-tenant auth / enterprise feature store

---

Live Demo

Surface	URL
Public lab	https://decision-lab-demo.vercel.app
GitHub	https://github.com/BarujaFe1/DecisionLab

How to try: load the 3 demo policies → compare expected cost → explore cost / sensitivity / audit trail → export the decision memo (.md).

---

Screenshots

Policy Builder	Scenario Simulator
Cost matrix	Threshold explorer
Fairness panel	Audit trail
Version comparison	Executive memo

---

Stack

Layer	Technology
Web	Next.js 15, React 19, TypeScript, Recharts, Lucide
Engine (browser)	TypeScript simulator in apps/web/lib/engine
API (optional)	FastAPI, Pandas, NumPy, SciPy, pytest

---

Architecture

apps/
  web/
    app/                 Next.js UI
    lib/engine/          policies, simulator, types (client-side)
    public/data/         synthetic ops decision cases
  api/
    app/services/        rules, simulator, demo_data (local API)
assets/                  icon, hero, screenshots

Flow: cases → policy evaluation → cost aggregation → sensitivity sweep → comparison → audit + memo.

---

Quick Start

Prerequisites: Node.js 20+, Python 3.10+ (optional for API), Git.

Frontend-only (same as Vercel)

cd apps/web
npm install
npm run dev

Windows integrated (web + API)

.\start.bat

FastAPI only

cd apps/api
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

---

Technical decisions

• Cost matrix over accuracy-only so false positives/negatives have prices
• Client-side engine on Vercel for a reliable lab without hosting SciPy
• Audit trail + memo to make the policy choice explainable before rollout
• Human review required — simulation is decision support, not auto-approve

---

Roadmap

• Policy versioning with named releases
• Richer subgroup fairness diagnostics
• Import of scored case CSVs
• Optional auth for shared lab workspaces
• Deeper coupling between web engine and FastAPI parity tests

---

Author

Felipe Alirio Baruja — data / product / full-stack portfolio.

• Portfolio: https://barujafe.vercel.app/
• GitHub: https://github.com/BarujaFe1
• LinkedIn: https://www.linkedin.com/in/barujafe/

---

License

MIT — see LICENSE.