Capstone project build and evidence pack

Your capstone project is to design, build, and document a complete AI agent system that solves a real-world problem. This is not a toy example. It should be something you could actually use or deploy.

Core Requirements (All projects must include):

Choose a project that interests you. Here are some suggestions organised by difficulty:

Foundation Level (Recommended for first-time builders):

1. Personal Research Assistant

   • Searches the web for information

   • Summarises findings

   • Saves notes to files

   • Tools: web_search, file_write, summarise

2. Email Draft Generator

   • Takes bullet points as input

   • Generates professional email drafts

   • Adjusts tone based on recipient type

   • Tools: tone_classifier, email_template, grammar_check

3. Code Documentation Bot

   • Reads Python files

   • Generates docstrings for functions

   • Creates README sections

   • Tools: file_read, docstring_generator, markdown_writer

Intermediate Level:

1. Customer Support Agent

   • Classifies incoming queries

   • Searches knowledge base for answers

   • Escalates complex issues to humans

   • Tools: classifier, knowledge_search, ticket_create, human_escalate

2. Data Analysis Assistant

   • Loads CSV files

   • Performs statistical analysis

   • Generates charts

   • Explains findings in plain English

   • Tools: csv_loader, calculator, chart_generator, explain

3. Meeting Summariser

   • Processes meeting transcripts

   • Extracts action items

   • Identifies decisions made

   • Creates structured summaries

   • Tools: text_splitter, entity_extractor, summarise, format_output

Advanced Level:

1. Multi-Agent Research Team

   • Supervisor coordinates specialists

   • Researcher finds information

   • Writer creates content

   • Critic reviews and improves

   • Tools: Multiple per agent, handoff between agents

2. Workflow Automation Agent

   • Monitors email for triggers

   • Executes automated responses

   • Logs all actions for audit

   • Supports human-in-the-loop for critical decisions

3. Security Assessment Agent

   • Analyses code for vulnerabilities

   • Checks configurations

   • Generates security reports

   • Prioritises findings by severity

4. MCP Server Ecosystem

• Build 3+ MCP servers connecting different data sources

• Create a unified client that discovers and uses them

• Implement OAuth 2.1 authentication

• Document the ecosystem with architecture diagrams

• Demonstrate tool annotations (read-only vs destructive)

Use this structure for your submission:

my-capstone-project/
├── README.md              # Project overview and setup
├── CLAUDE.md              # Context engineering documentation
├── ARCHITECTURE.md        # Design decisions and diagrams
├── SECURITY.md            # Security considerations and OWASP mapping
├── requirements.txt       # Python dependencies
├── src/
│   ├── __init__.py
│   ├── agent.py           # Main agent implementation
│   ├── mcp_server.py      # MCP server (if applicable)
│   ├── tools/
│   │   ├── __init__.py
│   │   ├── tool_1.py
│   │   ├── tool_2.py
│   │   └── tool_3.py
│   └── utils/
│       ├── __init__.py
│       ├── validation.py
│       └── logging.py
├── tests/
│   ├── test_agent.py
│   ├── test_tools.py
│   └── test_integration.py
└── examples/
    ├── example_1.py
    └── example_2.py

Your README.md should include:

# [Project Name]

## Overview
[One paragraph describing what your agent does and why it is useful]

## Architecture

[Include a Mermaid diagram showing your agent's structure]

## Features
- Feature 1
- Feature 2
- Feature 3

## Setup

### Prerequisites
- Python 3.10+
- Ollama installed and running

### Installation
```bash
pip install -r requirements.txt
```

### Configuration
[Any environment variables or config files needed]

## Usage

### Basic Example
```python
from src.agent import MyAgent

agent = MyAgent()
result = agent.run("Your query here")
print(result)
```

## Security Considerations
[Document your security measures and limitations]

## Testing
```bash
pytest tests/
```

## Limitations
[Be honest about what your agent cannot do]

## Future Improvements
[What would you add with more time?]

Your project will be evaluated on these criteria:

| Criteria | Weight | Description | |----------|--------|-------------| | Functionality | 25% | Does the agent work correctly? | | Code Quality | 20% | Clean, readable, well-organised code | | Security | 20% | OWASP Agentic risks addressed, least agency applied | | Documentation | 15% | Clear docs including context engineering notes | | Testing | 10% | Adequate test coverage | | Innovation | 10% | Creative problem-solving |

Grading Scale:

• Distinction (90%+): Exceptional work exceeding all requirements

• Merit (70-89%): Strong work meeting all requirements

• Pass (50-69%): Adequate work meeting minimum requirements

• Refer (Below 50%): Needs improvement before certification