19. Building an AI-enabled application

Start with a real problem

Pick a problem from real life. Keep it small. Some ideas:

Three ways to build, from easy to harder

Way 1: A custom chatbot (no code)

Make a chatbot with a fixed role and instructions.

• Gemini Gems: In Gemini, click Gems → New Gem. Give it a name, instructions, and optionally files. Save. Now it is your app.
• ChatGPT custom GPTs: Similar, under Explore GPTs → Create (may need a paid plan).
• NotebookLM: Upload your college handbook. Share the notebook. It is now a "campus helper" app.

Example instructions for a Gem:

You are "Vizag Campus Buddy", a helper for first year students of GVPIHLR. Answer questions about college life, study tips and Visakhapatnam in simple English. If someone asks about exam dates or fees, tell them to check the college website or ask the office. Be friendly. Keep answers under 100 words.

Way 2: A small web app with Google AI Studio (little or no code)

1. Open aistudio.google.com.
2. Go to the Build section.
3. Describe your app in one paragraph. Example: "A web app where a farmer uploads a photo of a plant leaf. The app says what the problem might be and gives 3 simple steps, in English and Telugu."
4. AI Studio writes the code and shows a working app. Try it. Ask for changes.
5. You can share the link or download the code.

Way 3: Write the code yourself with an API (some code)

An API lets your own program talk to the AI model. You need an API key from AI Studio (free).

# pip install google-genai
from google import genai

client = genai.Client(api_key="PASTE_YOUR_KEY_HERE")

topic = input("Which topic do you want explained? ")
response = client.models.generate_content(
    model="gemini-2.5-flash",
    contents=f"Explain {topic} to a 10th class student in Telugu. Under 100 words.",
)
print(response.text)

Run this in Google Colab or on your laptop. That is a complete AI app in 8 lines. Add a loop, a menu or a web page with Streamlit to make it bigger.

Design before you build

Write one page answering:

1. Problem: What problem? For whom?
2. Input: What does the user give? Text, photo, voice?
3. AI step: What does the model do?
4. Output: What does the user get?
5. Risks: What if the AI is wrong? Privacy? Bias? (Chapter 6)
6. Limits: What will the app not do? (A health app must say "see a doctor".)