Project: a streaming chat assistant

The briefCopy Anchor Link

A web page with a text box. The user sends text and sees the answer stream in. The assistant has one job set by a system prompt, for example: translate the text into the language the user picks and return only the translation.

StepsCopy Anchor Link

1. Make one request from a script and print the answer and the token usage.
2. Move the call to a backend route so the key stays on the server.
3. Keep a messages array so follow-up questions work.
4. Stream the answer to the page.
5. Handle 429 and 5xx responses with backoff and a friendly message.
6. Render markdown safely.
7. Add three few-shot examples to the system prompt and compare output before and after.

Check yourselfCopy Anchor Link

Ten test inputs including a very long one and one that tries to override the system prompt. Record what happens. Add a max output length and a per-user rate limit. Write down the cost of 100 requests from the token counts.

Done whenCopy Anchor Link

A stranger can open the page and translate text. The answer streams in word by word. A failed request (a wrong key, or the network off) shows a friendly message instead of a crash. The API key appears nowhere in the browser or the repository. Your ten test inputs and what happened are written in the README.

StretchCopy Anchor Link

Add a temperature control and see how it changes translations. Pin a model snapshot and write the one line in the README that says why.

StarterCopy Anchor Link

Use Python with the Anthropic SDK (pip install anthropic, set ANTHROPIC_API_KEY). TypeScript works the same way. Start with a terminal version, then move it behind a web route. Save this as chat.py:

from anthropic import Anthropic

client = Anthropic()
SYSTEM = "Translate the user's text into the language they name. Return only the translation."
history = []

while True:
    text = input("you> ")
    history.append({"role": "user", "content": text})
    pieces = []
    with client.messages.stream(model="claude-sonnet-5-5", max_tokens=400, system=SYSTEM, messages=history) as stream:
        for piece in stream.text_stream:
            print(piece, end="", flush=True)
            pieces.append(piece)
    answer = "".join(pieces)
    print()
    history.append({"role": "assistant", "content": answer})

Run it with python chat.py. It already streams and keeps history. Your work is steps 2 to 7: the web route, error handling, the safe rendering, the tests.