AI Engineering — Setup Guide
What You Need to Get Started
AI engineering doesn't require installing a local model. You use API access to cloud models (Claude, GPT-4, etc.) and Python libraries. Here's everything to set up.
Step 1: Python Setup
bash
# Verify Python 3.10+ is installed
python3 --version # Need 3.10 or higher
# Create a virtual environment for AI projects
mkdir ai-projects && cd ai-projects
python3 -m venv venv
source venv/bin/activate # Mac/Linux
# venv\Scripts\activate # Windows
# Verify
which python # should point to venv/bin/python
Step 2: Get an Anthropic API Key (Claude)
1.Go to console.anthropic.com
2.Create account → API Keys → Create Key
3.Copy the key (starts with sk-ant-...)
bash
# Set as environment variable (don't hardcode in code!)
export ANTHROPIC_API_KEY="sk-ant-your-key-here"
# Add to ~/.bashrc or ~/.zshrc to persist
echo 'export ANTHROPIC_API_KEY="sk-ant-your-key-here"' >> ~/.bashrc
# Test the key
pip install anthropic
python3 -c "
import anthropic
client = anthropic.Anthropic()
msg = client.messages.create(
model='claude-haiku-4-5-20251001',
max_tokens=100,
messages=[{'role': 'user', 'content': 'Say hello!'}]
)
print(msg.content[0].text)
"
# Should print a greeting
Step 3: Install Core AI Libraries
bash
# Anthropic SDK (Claude)
pip install anthropic
# OpenAI SDK (GPT-4, also used with many other providers)
pip install openai
# LangChain (chain LLM calls, build RAG systems)
pip install langchain langchain-community langchain-anthropic
# Vector database for RAG
pip install chromadb
# Document loaders
pip install pypdf python-docx
# Embeddings (free, runs locally)
pip install sentence-transformers
# Data processing
pip install pandas numpy
# Web interface for demos
pip install streamlit
# Save all dependencies
pip freeze > requirements.txt
Step 4: Test Everything Works
python
# test_setup.py
import anthropic
import langchain
import chromadb
print("All imports successful!")
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-haiku-4-5-20251001",
max_tokens=50,
messages=[{"role": "user", "content": "What is 2+2?"}]
)
print(f"Claude says: {response.content[0].text}")
print(f"Tokens used: {response.usage.input_tokens} in, {response.usage.output_tokens} out")
print(f"Cost: ~${(response.usage.input_tokens * 0.00025 + response.usage.output_tokens * 0.00125) / 1000:.5f}")
bash
python3 test_setup.py
# Claude says: 2+2 equals 4.
# Tokens used: 14 in, 12 out
# Cost: ~$0.00002
API Costs — What to Expect
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Good for |
|---|
|---|---|---|---|
| claude-haiku-4-5 | $0.25 | $1.25 | Q&A, classification, summaries |
|---|
| claude-sonnet-4-6 | $3.00 | $15.00 | Complex tasks, code, analysis |
| claude-opus | $15.00 | $75.00 | Most complex reasoning |
Rule: Start with Haiku. Use Sonnet for tasks Haiku struggles with. 1000 Haiku API calls costs roughly ₹20-50. Very affordable for learning.
Optional: Run Models Locally (Free, No API Cost)
bash
# Install Ollama (runs open-source models locally)
# macOS/Linux:
curl -fsSL https://ollama.ai/install.sh | sh
# Pull a model (llama3.2 is good and small)
ollama pull llama3.2
# Run it
ollama run llama3.2
# You can now chat with it in the terminal — no API key, no cost
# Use from Python
pip install ollama
python3 -c "
import ollama
r = ollama.chat(model='llama3.2', messages=[{'role':'user','content':'Hello!'}])
print(r['message']['content'])
"
Project Structure
ai-projects/
├── venv/ # Virtual environment (gitignored)
├── .env # API keys (NEVER commit this!)
├── requirements.txt # Dependencies
├── 01_hello_claude.py # Basic API call
├── 02_chat_history.py # Multi-turn conversation
├── 03_rag_system.py # RAG with documents
└── data/ # Your documents for RAG
bash
# .env file (use python-dotenv to load)
ANTHROPIC_API_KEY=sk-ant-xxx
OPENAI_API_KEY=sk-xxx # If using OpenAI too
Next Steps
Your AI development environment is ready. Go to Fundamentals to start building with the Claude API.
