Can I vibe code Sift?

Build a real-time risk scoring and fraud decisioning API microservice designed for e-commerce checkout and account events using Node.js (TypeScript), Fastify, PostgreSQL, and Redis.

DATA MODEL & ENTITIES:
- Entity: User (id, email, phone, created_at, baseline_risk_score)
- Entity: DeviceSession (id, user_id, fingerprint_hash, ip_address, user_agent, raw_signals_json, created_at)
- Entity: Event (id, event_type ['$transaction', '$create_account', '$login'], user_id, session_id, payload_json, timestamp)
- Entity: Rule (id, name, condition_expression, action ['APPROVE', 'DENY', 'WORKFLOW_REVIEW'], priority, active)
- Entity: FraudScore (id, event_id, calculated_score [0.00 to 1.00], triggers_json, decision, execution_time_ms)

CORE PIPELINE & INTEGRATIONS:
1. REST API endpoint POST /v1/events accepting standard Sift-like payloads (event_type, user_id, payment_method, billing_address, shipping_address, amount, currency, session_id).
2. Integrate client-side device fingerprinting signal ingestion (parse client IP, query MaxMind GeoIP2 database for proxy/VPN detection, compute risk delta).
3. Rule Engine: Execute a chain of deterministic rules evaluated via a sandboxed JS evaluator (e.g., condition: "amount > 1000 AND ip_country != billing_country" => DENY).
4. Machine Learning Inference Mock/Pipeline: Implement a ONNX Runtime or local XGBoost microservice caller passing normalized vector features (email domain age, distance between billing/shipping zip, velocity of transactions per user in Redis last 1 hour) returning a probability score between 0 and 1.
5. Decision Engine: Combine rule overrides and ML threshold scoring to output a final decision: APPROVE (score < 0.3), REVIEW (0.3 <= score < 0.7), DENY (score >= 0.7).
6. Webhooks: Post score decision async to configured Shopify or platform webhooks.

FAILURE MODES & RESILIENCE:
- Latency strictly under 100ms. Async offload of logging and heavy analytics to background Redis/BullMQ workers.
- If the ML inference engine times out (>50ms), fallback gracefully to static Rule Engine decisions with a flag `fallback_used: true`.

OUT OF SCOPE:
- Chargeback liability guarantees or insurance reimbursement workflows.
- Native mobile SDK development (REST API ingestion only).

$ each button prefixes agent-specific run instructions · build your own product, never copy proprietary code, trademarks or designs