Real-World Project: Building an AI Tooling Stack

Introduction

Every lesson so far covered one tool in isolation. Real projects combine several of them. This final lesson wires four categories from across the course into one complete, working support assistant — retrieval, generation, cost tracking, and a guardrail against making things up — closing the course with the whole stack working together instead of one piece at a time.

A Real-Life Analogy to Close the Course

Remember lesson 4's toolbox: a hammer, a screwdriver, a wrench, each suited to a different job. This whole course has been about learning what is in that toolbox, one tool at a time. But nobody builds an actual piece of furniture using exactly one tool — a real project reaches for the screwdriver for the screws, the hammer for the nails, and the tape measure to make sure everything lines up, all in the same afternoon. This lesson is that afternoon: taking several specific tools you've now individually learned and using them together on one real, finished thing.

The Project: A Support Assistant

The scenario: a small company wants a chatbot that answers customer questions using their existing support documentation, refuses to guess when the answer isn't in that documentation, and reports what each answer cost — the same "chat with your docs" pattern from lesson 3's examples, now assembled end to end as a real project.

Assembling the Stack

Complete Code

from dotenv import load_dotenv
import chromadb
from openai import OpenAI

load_dotenv()

client_openai = OpenAI()
client_chroma = chromadb.PersistentClient(path="./chroma-db")
collection = client_chroma.get_or_create_collection("support-docs")

collection.add(
    ids=["doc-1", "doc-2", "doc-3"],
    documents=[
        "Refunds are processed within 5-7 business days of approval.",
        "Shipping takes 3-5 business days for domestic orders, 10-14 for international.",
        "Our support team is available Monday-Friday, 9am-6pm ET.",
    ],
)

COST_PER_1K_INPUT = 0.00015
COST_PER_1K_OUTPUT = 0.0006

def ask_support_assistant(question: str) -> dict:
    results = collection.query(query_texts=[question], n_results=2)
    context = "\n".join(results["documents"][0])

    response = client_openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {
                "role": "system",
                "content": "Answer only using the provided context. If the answer isn't in the context, say you don't know and suggest contacting support.",
            },
            {"role": "user", "content": f"Context:\n{context}\n\nQuestion: {question}"},
        ],
    )

    usage = response.usage
    cost = (usage.prompt_tokens / 1000) * COST_PER_1K_INPUT + (usage.completion_tokens / 1000) * COST_PER_1K_OUTPUT

    return {
        "answer": response.choices[0].message.content,
        "cost": round(cost, 6),
    }

result = ask_support_assistant("How long until my order arrives?")
print(result["answer"])
print(f"Cost: ${result['cost']}")

result2 = ask_support_assistant("Do you offer price matching?")
print(result2["answer"])

Common Mistakes Across the Whole Stack

Best Practices Across the Whole Stack

Frequently Asked Questions

Key Takeaways

Summary

Across 24 lessons, this course moved from a single API call to a complete, multi-category AI application stack. The tools will keep changing — new frameworks, new models, new categories — but the underlying patterns you have now seen in action (provider SDKs, retrieval, orchestration, evaluation, cost awareness) are what will keep transferring to whatever comes next.