EdgeLab System Architecture ┌─────────────────────────────────────────────────────────────────┐
│ USER │
│ (Browser/Client) │
└───────────────────────────┬─────────────────────────────────────┘
│
│ HTTP Requests
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ STREAMLIT UI (Port 8501) │
│ ┌────────────┬──────────────┬────────────────┐ │
│ │ Browse │ Submit │ View │ │
│ │ Problems │ Solution │ Results │ │
│ └────────────┴──────────────┴────────────────┘ │
└───────────────────────────┬─────────────────────────────────────┘
│
│ REST API Calls
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ FLASK API (Port 5000) │
│ │
│ Endpoints: │
│ - GET /assignments (list problems) │
│ - GET /assignments/:id (get problem details) │
│ - POST /submit (submit code) │
│ - GET /results/:id (get submission results) │
│ │
└─────┬──────────────────────┬──────────────────────┬─────────────┘
│ │ │
│ │ │
▼ ▼ ▼
┌──────────────┐ ┌──────────────────┐ ┌─────────────────────┐
│ DATABASE │ │ CODE EXECUTOR │ │ AI FEEDBACK │
│ (SQLite) │ │ │ │ (Phi-3/Ollama) │
│ │ │ ┌────────────┐ │ │ │
│ - assignments│ │ │ Docker │ │ │ - Analyze code │
│ - submissions│ │ │ Containers │ │ │ - Generate tips │
│ - results │ │ │ │ │ │ - Only public info │
│ │ │ │ Python:3.11│ │ │ │
│ │ │ │ OpenJDK:11 │ │ │ (Optional) │
│ │ │ │ Alpine+SQL │ │ │ │
│ │ │ └────────────┘ │ │ │
└──────────────┘ └──────────────────┘ └─────────────────────┘
User browses assignments via UI
UI requests assignment list from API
API queries database, strips hidden tests, returns public data
User writes code and clicks "Submit"
UI sends code to API /submit endpoint
API:
Saves submission to database (status: pending)
Spawns background thread for execution
Returns submission ID immediately
Background thread:
Retrieves full assignment (with hidden tests) from DB
Calls executor with code and tests
Executor spawns isolated Docker container
Runs code with timeouts and resource limits
Compares output to expected results
Returns test results
(Optional) Calls AI for feedback
Updates submission in DB with results
UI polls /results endpoint
API returns results (hiding hidden test details)
UI displays score, test breakdown, and feedback
┌─────────────────────────────────────────────────────────────┐
│ LAYER 1: Network Isolation │
│ - All containers run with --network none │
│ - No outbound connections possible │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│ LAYER 2: Resource Limits │
│ - Memory cap: 128-256 MB │
│ - CPU limit: 0.5 cores │
│ - Execution timeout: 5 seconds │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│ LAYER 3: Filesystem Isolation │
│ - Code mounted read-only │
│ - Temp files only │
│ - No persistent storage │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│ LAYER 4: Information Hiding │
│ - Hidden tests never sent to frontend │
│ - API strips sensitive data before responses │
│ - Results show pass/fail count only for hidden tests │
└─────────────────────────────────────────────────────────────┘
Data Flow: Submission Lifecycle [User Code]
│
▼
[API: Create Submission]
│ (Save to DB with status='pending')
▼
[Background Thread Started]
│
▼
[Executor: Get Full Assignment]
│ (Including hidden tests)
▼
[For Each Test Case:]
│
├─► [Spawn Docker Container]
│ │
│ ├─► [Copy code to container]
│ ├─► [Set resource limits]
│ ├─► [Execute with timeout]
│ ├─► [Capture output]
│ └─► [Destroy container]
│
└─► [Compare output vs expected]
│
▼
[Record pass/fail]
[All Tests Complete]
│
▼
[Calculate Score]
│
▼
[Optional: Get AI Feedback]
│
▼
[Update DB: status='passed/failed', results, feedback]
│
▼
[User polls /results]
│
▼
[API: Return sanitized results]
│ (No hidden test details)
▼
[UI: Display results]
Component Responsibilities
Present problems in readable format
Provide code editor interface
Handle user interactions
Poll for results
Display test outcomes
Route HTTP requests
Validate inputs
Coordinate components
Manage async execution
Enforce security policies
Store persistent data
Provide data access layer
Handle migrations
Load assignments from files
Spawn isolated environments
Enforce resource limits
Run code safely
Compare outputs
Handle errors gracefully
Analyze code quality
Generate constructive tips
Operate on public data only
Fallback to simple feedback
Component
Technology
Why?
Frontend
Streamlit
Fast to build, looks good
API
Flask
Simple, Python-native
Database
SQLite
No setup, sufficient for demo
Containers
Docker
Industry standard, secure
AI
Phi-3 + Ollama
Local, free, good code reasoning
Language
Python 3.11
Familiar, good libraries
$ docker compose up
Docker Compose orchestrates:
- Builds API image (Flask + executor)
- Builds UI image (Streamlit)
- Creates network bridge
- Mounts volumes (database, assignments)
- Exposes ports (5000, 8501)
- Starts services in dependency order
/home/claude/edgelab/
│
├── api/ # Backend service
│ ├── app.py # Flask routes
│ ├── database.py # DB operations
│ ├── executor.py # Code runner
│ ├── ai_feedback.py # AI integration
│ └── requirements.txt
│
├── ui/ # Frontend service
│ ├── streamlit_app.py # UI logic
│ └── requirements.txt
│
├── assignments/ # Problem definitions
│ ├── python_two_sum.json
│ ├── sql_top_salaries.json
│ └── java_palindrome.json
│
├── database/ # Persistent storage
│ └── edgelab.db # SQLite file
│
├── docker-compose.yml # Orchestration
├── Dockerfile.api # API image build
├── Dockerfile.ui # UI image build
└── README.md # Documentation