✈️ Travel Suggestion

GPT-4o mini via OpenAI

Concept

This demo asks an LLM to suggest one thing to do in a city you name. It uses OpenAI's GPT-4o mini because the task is open-ended creative writing that benefits from strong instruction-following, while staying cheap and fast enough for an interactive demo.

A short system prompt ("You are a witty travel guide") gives the model a consistent persona for every request, and the user prompt asks for exactly one suggestion so the reply stays short enough to fit in the result card.

Theory & Concepts

Prompt Engineering for UI Constraints

The Travel Suggestion demo illustrates how careful prompt design can adapt LLM outputs to fit specific application requirements:

  • Enforcing Brevity: In a UI designed around compact result cards, long-winded answers break the layout. By explicitly structuring the user prompt as "Suggest one thing to do in {city}", we force the GPT-4o mini model to narrow its focus. This guarantees a concise, punchy response that fits perfectly into the demo's visual design.
  • Stateless Persona Maintenance: Every time you click "Suggest something," the backend makes a fresh, isolated API call. Because the OpenAI API is stateless, it doesn't remember your previous city searches. Therefore, we must inject the system prompt ("You are a witty travel guide.") into every single request to guarantee the tone remains consistently playful, regardless of how many times you use the tool.

Request flow

🧑 Browser Type a city, click “Suggest something”
→
POST /travel Flask route
→
get_travel_suggestion() Builds the prompt
→
OpenAI API GPT-4o mini suggests an activity
→
🧑 Browser Suggestion rendered in the result box

Code flow

flowchart TD A[Browser
city] -->|POST /travel| B[app.py
travel route] B -->|city| C[travel.py
get_travel_suggestion] C -->|prompt| D[OpenAI API
GPT-4o mini] D -->|suggestion| C C -->|suggestion| B B -->|JSON result| A

Backend

Generates the suggestion via OpenAI's chat completions API. 
from config import get_openai_client

# gpt-4o-mini: fast, cheap, and good enough for a short creative reply.
TRAVEL_MODEL = "gpt-4o-mini"


def get_travel_suggestion(city: str = "Bangalore") -> str:
    client = get_openai_client()
    response = client.chat.completions.create(
        model=TRAVEL_MODEL,
        messages=[
            {
                "role": "system",
                # A brief persona keeps every response short and fun.
                "content": "You are a witty travel guide.",
            },
            {
                "role": "user",
                # Asking for exactly one suggestion keeps the reply concise.
                "content": f"Suggest one thing to do in {city}.",
            },
        ],
    )
    return response.choices[0].message.content

API route

Exposes the travel suggester over HTTP as POST /travel. 
@bp.route("/travel", methods=["POST"])
def travel():
    data = request.get_json(force=True)
    # Fall back to Bangalore so the demo always has a usable city.
    city = (data.get("city") or "").strip() or "Bangalore"
    try:
        text = get_travel_suggestion(city)
        return jsonify({"result": text})
    except Exception as e:
        return jsonify({"error": str(e)}), 500