CPSC 436C - AI Development Environment Setup

🎯 Your Goal

Yyou should be able to:

Run AI models locally (not just web interfaces). Note: "locally" just means a machine you control.
Compare responses across different models
Document your AI interactions for course projects
Critically evaluate AI outputs for technical accuracy

⚠️ Important Notes

Time Investment: Plan 1-2 hours for initial setup
Disk Space: You'll need 5-10GB free space for models
Internet: Initial model downloads can be large (1-4GB each)
Help Available: TA office hours specifically for setup support

1 Python Environment Setup

First, let's get a modern Python environment. We recommend uv (modern) but pip works too.

Using uv (Recommended)

# Install uv

curl -LsSf https://astral.sh/uv/install.sh | sh

# Create project directory

mkdir ~/cpsc436c-ai

cd ~/cpsc436c-ai

# Initialize Python project

uv init

uv add litellm

Using pip (Traditional)

# Create project directory

mkdir ~/cpsc436c-ai

cd ~/cpsc436c-ai

# Create virtual environment

python -m venv .venv

source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install requirements

pip install litellm

✅ Verification: Run python -c "import litellm; print('Success')"

2 Local AI Models Setup

Choose ONE of these options to run models locally:

Option A: Ollama

Download from ollama.ai
Install and start Ollama
Download a model:

# Start with a smaller model for testing

ollama pull llama3.2:3b

# Or a more capable model (larger download)

ollama pull llama3.1:8b

Option B: LM Studio

Download from lmstudio.ai
Install and launch LM Studio
Browse and download a model (try llama-3.2-3b-instruct for starting)
Start the local server in LM Studio

✅ Verification: You should be able to chat with your local model through the respective interface.

3 Model Comparison Setup

Set up litellm proxy to compare different models:

# Create a config file

cat > config.yaml << EOF

model_list:

  - model_name: local-llama

    litellm_params:

      model: ollama/llama3.2:3b

      api_base: http://localhost:11434

  - model_name: gpt-5

    litellm_params:

      model: gpt-5

      # Add your OpenAI API key if you have one

  - model_name: claude-sonnet

    litellm_params:

      model: claude-sonnet-4-20250514

      # Add your Anthropic API key if you have one
EOF

# Start the proxy

litellm --config config.yaml

✅ Verification: Visit http://localhost:4000 to see the litellm interface.

4 Test Your Setup

Create a simple test to verify everything works:

# Create test.py

cat > test.py << EOF

import litellm

# Test local model

response = litellm.completion(

    model="ollama/llama3.2:3b",

    messages=[{"role": "user", "content": "Explain serverless vs containers in 2 sentences"}],

    api_base="http://localhost:11434"

)

print("Local model response:", response.choices[0].message.content)

EOF

# Run the test

python test.py

✅ Verification: You should get a reasonable 2-sentence explanation about serverless vs containers.

5 Document Your First AI Collaboration

For Thursday's class, try this exercise:

Ask your local model: "What's the best database for my project?"
Note the generic response you get
Ask: "I need a database for unpredictable traffic, AWS free tier, fast key-value lookups for a URL shortener. Compare DynamoDB vs RDS Aurora Serverless with cost and performance trade-offs."
Compare the quality of responses
Write 2-3 sentences about what you learned

✅ Verification: Bring your documented comparison to Thursday's class.

🆘 Need Help?

TA Office Hours: Dedicated AI setup support sessions
Discord: #setup channel for setup questions
Common Issues: Check pinned messages in Discord for solutions
Alternative: If local setup fails, you can use web interfaces temporarily, but document what didn't work

Remember: The goal is professional AI collaboration skills, not perfect technical setup. If you're struggling, document the problems - that's valuable learning too!