Use Cases

Real-world scenarios where Purple8 Builder delivers a complete, working codebase — not a template, not a scaffold. Each example shows the input, the agents activated, and the output produced.

1. Multi-Tenant SaaS Invoicing Platform

Prompt:

"Build a multi-tenant SaaS invoicing app. Each tenant has their own clients, invoices, and branding. Include Stripe subscription billing, PDF export, a client portal, and role-based user management."

Goal: MVP · Mode: PLUS · HITL: enabled

Agents activated

1. IdeationAgent → Full requirements with tenant isolation model
2. ArchitectureAgent → Multi-tenant architecture (schema-per-tenant vs row-level security)
3. DatabaseAgent → PostgreSQL schema with tenant, invoice, line_item, client tables + RLS policies
4. APIDevelopmentAgent → REST API: /tenants, /invoices, /clients, /payments
5. BackendDevelopmentAgent → FastAPI service, Stripe webhook handler, PDF renderer (WeasyPrint)
6. FrontendDevelopmentAgent → Vue 3 dashboard: invoice builder UI, client portal, Stripe Checkout
7. InfrastructureDevelopmentAgent → Docker Compose, env templates, nginx config
8. SecurityQAAgent → Stripe key handling review, CSRF protection, tenant isolation audit
9. UnittestAgent → pytest suite for invoice calculation, Stripe mock, PDF generation
10. DeploymentAgent → GitHub Actions CI/CD, Railway deploy config

Output

generated_projects/saas-invoicing/
├── backend/
│   ├── app/
│   │   ├── routers/invoices.py, clients.py, payments.py
│   │   ├── models/invoice.py, tenant.py
│   │   ├── services/stripe_service.py, pdf_service.py
│   │   └── middleware/tenant_middleware.py
│   └── tests/test_invoices.py, test_stripe.py
├── frontend/
│   ├── src/views/InvoiceBuilder.vue, ClientPortal.vue
│   └── src/stores/invoiceStore.ts
├── migrations/001_initial.sql
├── docker-compose.yml
├── .github/workflows/ci.yml
└── README.md

Build time: ~12 minutes · Estimated LLM cost: $0.84

---

2. AI-Powered Resume Screening API

Prompt:

"Build an API that accepts a job description and a batch of résumés (PDF/DOCX), extracts structured data from each, scores them against the job description using an LLM, ranks candidates, and returns a structured JSON report."

Goal: Production · Mode: PLUS

Agents activated

1. IdeationAgent → Requirements: async batch processing, structured extraction, ranking rubric
2. DatabaseAgent → Schema: jobs, candidates, scores, processing_jobs tables
3. RAGAgent → Vector store for job description embeddings and résumé chunks
4. DocumentIntelligenceAgent → PDF/DOCX parser, section extractor (Experience, Skills, Education)
5. NLPAgent → Skill taxonomy normalisation, keyword extraction
6. APIDevelopmentAgent → REST API: POST /jobs, POST /jobs/{id}/screen, GET /jobs/{id}/results
7. BackendDevelopmentAgent → FastAPI + Celery workers, LLM scoring pipeline
8. MLOpsAgent → MLflow experiment tracking for scoring model evaluation
9. EvaluationMetricsAgent → Accuracy benchmarks against labelled test set
10. SecurityQAAgent → PII handling, data retention policy enforcement
11. UnittestAgent → Tests for parser, scorer, ranking logic
12. APITestAgent → Contract tests for all endpoints
13. ObservabilityAgent → Prometheus metrics: queue depth, scoring latency, throughput

Key generated files

• services/screener/scoring_pipeline.py — LLM-based relevance scoring with rubric
• services/screener/document_parser.py — multi-format extraction
• workers/screening_worker.py — Celery task with retry logic
• openapi.json — full OpenAPI 3.1 spec

---

3. Real-Time Logistics Dashboard

Prompt:

"Build a real-time dashboard that ingests GPS telemetry from delivery vehicles via MQTT, shows live positions on a map, calculates ETA using traffic data, and alerts dispatchers when vehicles deviate from route by more than 500m."

Goal: MVP · Mode: PLUS

Agents activated

1. IdeationAgent → Requirements: MQTT ingestion, geofence alerting, ETA calculation
2. ArchitectureAgent → Event-driven: MQTT → RabbitMQ → consumer → PostgreSQL + Redis pubsub → WebSocket → frontend
3. DatabaseAgent → Schema: vehicles, telemetry, routes, alerts, geofences; PostGIS extension
4. DataEngineeringAgent → MQTT consumer, telemetry pipeline, route deviation calculator
5. BackendDevelopmentAgent → FastAPI + WebSocket server, ETA engine (OSRM integration)
6. FrontendDevelopmentAgent → Vue 3 + Leaflet.js map dashboard, live vehicle markers, alert panel
7. TimeSeriesAgent → Telemetry downsampling, anomaly detection for stale GPS
8. DashboardAgent → Grafana dashboard config for vehicle metrics
9. LoadTestAgent → k6 script: 500 concurrent vehicle streams
10. InfrastructureDevelopmentAgent → Docker Compose with Mosquitto MQTT broker, PostGIS
11. ObservabilityAgent → Structured logging for every telemetry event, alert audit trail

What makes this non-trivial

Builder detected the real-time geospatial domain and activated DataEngineeringAgent + TimeSeriesAgent without being told — the knowledge graph matched "GPS telemetry" and "real-time" to those specialist agents automatically.

---

4. Internal Developer Portal

Prompt:

"Build a self-service developer portal where engineers can provision new microservices from templates, manage their GitHub repos, view CI/CD status, and rotate secrets — all from one UI. Must integrate with our existing GitHub org and Vault."

Goal: Production · HITL: on all phases

Agents activated

1. IdeationAgent → Feature map: service provisioning, repo management, secret rotation, CI status
2. ArchitectureAgent → Hub-and-spoke: portal → GitHub API, Vault API, Kubernetes API
3. APIDevelopmentAgent → Internal REST API with service mesh auth (mTLS)
4. EnterpriseIntegrationAgent → GitHub App integration, Vault dynamic secrets, Tekton/ArgoCD webhooks
5. BackendDevelopmentAgent → Python service, GitHub App SDK, hvac (Vault client)
6. FrontendDevelopmentAgent → React portal: service catalog, provisioning wizard, secret manager
7. EnterpriseCyberSecurityAgent → Vault policy review, least-privilege GitHub App scopes, OIDC flow
8. ComplianceAgent → SOC 2 control documentation, change management audit trail
9. AuditAgent → Immutable audit log for all provisioning and secret rotation events
10. DevOpsAgent → Helm chart for portal deployment, ArgoCD application manifest
11. E2ETestAgent → Playwright: end-to-end service provisioning flow
12. DocumentIntelligenceAgent → Auto-generated runbooks from provisioning templates

Notable outputs

• Complete Vault policy files for dynamic secret backends
• GitHub App manifest with minimum-required scopes
• SOC 2 control narrative document ready for auditor review

---

5. Healthcare Patient Intake Chatbot

Prompt:

"Build a HIPAA-compliant patient intake chatbot for a GP clinic. It collects demographics, symptoms, and medical history, validates responses, and creates a structured pre-consultation note for the doctor. Needs to handle both English and Spanish."

Goal: Production · Mode: PLUS · HITL: enabled

Agents activated

1. IdeationAgent → Requirements with HIPAA constraints flagged, bilingual scope defined
2. NLPAgent → Intent classification, entity extraction (symptoms, medications, allergies)
3. AgenticWorkflowAgent → Multi-turn dialogue state machine: greeting → demographics → symptoms → history → confirmation
4. DatabaseAgent → Schema: patients, intake_sessions, pre_consult_notes; field-level encryption for PHI
5. BackendDevelopmentAgent → FastAPI + WebSocket chatbot server
6. FrontendDevelopmentAgent → Embedded chat widget (Vue 3 Web Component)
7. ComplianceAgent → HIPAA BAA template, PHI data map, retention policy
8. EnterpriseCyberSecurityAgent → TLS enforcement, PHI encryption at rest and in transit, access logging
9. AccessibilityQAAgent → WCAG 2.1 AA: keyboard navigation, screen reader labels, colour contrast
10. AuditAgent → Audit trail for every PHI access event

HIPAA-specific outputs

• Field-level AES-256 encryption for all PHI fields in the database schema
• Audit log schema meeting 45 CFR § 164.312 requirements
• BAA template document
• Data retention policy (6-year default for medical records)

---

6. E-Commerce Platform with AI Recommendations

Prompt:

"Build a full e-commerce platform: product catalogue, shopping cart, Stripe checkout, order management, and an AI recommendation engine that suggests products based on browsing history and purchase patterns."

Goal: Production · Mode: PLUS

Agents activated

1. IdeationAgent → Full e-commerce feature map
2. ArchitectureAgent → Microservices: catalogue-service, cart-service, order-service, recommendation-service
3. DatabaseAgent → Schema: products, orders, order_items, cart, users, browsing_events; vector column for product embeddings
4. RecommenderAgent → Collaborative filtering + content-based hybrid; product embedding pipeline using Purple8 Graph
5. APIDevelopmentAgent → REST + GraphQL API (Strawberry)
6. FrontendDevelopmentAgent → Vue 3 storefront: product listing, PDP, cart, checkout
7. BackendDevelopmentAgent → All four microservices + Stripe integration
8. InfrastructureDevelopmentAgent → Docker Compose, Nginx routing between services
9. PerformanceQAAgent → Load test: 1,000 concurrent shoppers, checkout latency < 200ms p95
10. LoadTestAgent → k6 load test script with checkout scenario
11. VisualRegressionAgent → Percy snapshots for product listing, PDP, cart
12. SecurityQAAgent → PCI DSS checklist for Stripe integration, CSP headers
13. SEOAgent → Server-side rendering config, structured data (JSON-LD), sitemap

Recommendation engine details

The RecommenderAgent generates a two-stage pipeline:

1. Candidate generation — ANN search over product embeddings (using Purple8 Graph hybrid search)
2. Ranking — LightGBM ranking model trained on synthetic click data during build

Total output: 312 files, 18,400 lines of code

---

Common Patterns

Domain detection is automatic

You don't need to tell Builder which agents to use. The knowledge graph detects domain signals in your prompt:

Signal in prompt	Agents added automatically
"HIPAA", "PHI", "medical"	ComplianceAgent, EnterpriseCyberSecurityAgent, AuditAgent
"ML", "model", "training"	MachineLearningAgent, MLOpsAgent, EvaluationMetricsAgent
"real-time", "telemetry", "MQTT"	DataEngineeringAgent, TimeSeriesAgent
"recommendation", "collaborative filtering"	RecommenderAgent, GraphAnalyticsAgent
"chatbot", "dialogue", "NLU"	NLPAgent, AgenticWorkflowAgent
"microservices", "Kubernetes"	InfrastructureDevelopmentAgent, DevOpsAgent

HITL is most valuable at phase boundaries

Use HITL checkpoints on planning to review architecture before code is written, and on release to review the full output before deployment. Skip HITL for prototypes.

Memory compounds across builds

If you've previously built a PostgreSQL + FastAPI project, the next build targeting the same stack reuses the established naming conventions, schema patterns that passed QA, and successful prompt strategies. Subsequent builds of similar projects are measurably faster and produce higher CTQ scores.