18. Full-Stack AI App Project

This capstone consolidates skills from the entire course. The project deliverables demonstrate production-ready development practices. Each deliverable has specific requirements and acceptance criteria.

Project Requirements

Build a document Q&A application with the following components:

1. Model Serving Layer

• Deploy llama.cpp or vLLM server
• Expose /completion and /embeddings endpoints
• Health check that verifies model loads
• Document startup time and memory usage

2. Backend API (FastAPI)

• POST /api/v1/upload - Upload PDF with progress tracking
• GET /api/v1/documents - List user documents
• GET /api/v1/documents/{id} - Get document status
• POST /api/v1/ask - Stream question response
• GET /api/v1/history - Get conversation history
• Authentication with JWT
• Input validation with Pydantic
• Structured logging with trace IDs

3. Frontend (React)

• Document upload with drag-drop and progress
• Document list with status indicators
• Chat interface with streaming responses
• Conversation history panel
• Error handling with retry actions

4. Infrastructure

• Docker Compose file with all services
• Health checks and dependencies
• Resource limits on all containers
• Secrets management

5. CI/CD Pipeline

• GitHub Actions workflow
• Lint, test, build stages
• Docker image push to registry
• Deployment to staging on merge

6. Monitoring

• Prometheus metrics endpoint
• Grafana dashboard JSON
• Alert rules for latency, errors, queue

Acceptance Criteria

The project must pass these checks:

1. Functional test: Upload a PDF, ask a question, receive an answer
2. Load test: Handle 10 concurrent users without errors
3. Security scan: Zero high-severity vulnerabilities in ZAP scan
4. Documentation: README, API docs, user guide, runbook
5. Monitoring: Dashboards show real metrics, alerts fire on test

Submission Format

ai-doc-qa/
├── backend/
│   ├── app/
│   ├── tests/
│   ├── Dockerfile
│   └── requirements.txt
├── frontend/
│   ├── src/
│   ├── Dockerfile
│   └── package.json
├── model-server/
│   ├── Dockerfile
│   └── start.sh
├── docker-compose.yml
├── prometheus.yml
├── grafana/
│   └── dashboards/
│       └── ai-app.json
├── nginx/
│   └── nginx.conf
├── docs/
│   ├── README.md
│   ├── api.md
│   ├── user-guide.md
│   └── runbook.md
├── k6/
│   └── load_test.js
└── .github/
    └── workflows/
        └── ci-cd.yml

Evaluation Rubric

Category	Points	Criteria
Code Quality	20	Clean structure, type hints, tests
Architecture	20	Clear separation, documented decisions
Deployment	20	Docker setup, CI/CD pipeline
Monitoring	15	Metrics, dashboards, alerting
Security	15	Input validation, auth, secrets
Documentation	10	Complete, accurate, usable

Next Steps

After completing the capstone:

1. Add features: document summarization, multi-modal support, multi-user workspaces
2. Optimize performance: caching embeddings, model quantization tuning
3. Scale infrastructure: Kubernetes migration, multi-region deployment
4. Improve reliability: chaos engineering, automated failover

The application serves as a foundation. Real production systems evolve based on user feedback and operational experience. Ship early, measure everything, iterate constantly.