chatbot

This is for explanation of chatbot, and quick integration for ur projects.

1. we install using pip install google-generativeai

2. get api key from https://aistudio.google.com/prompts/new_chat replace this actual key in api_key from actual code.

3. You can fine tune ur ai model using these props temperature : creative(1.0-less accurate), focused(0.2- less diverse), balamced(0.4-0.7) top_p(nucleus sampling) : high diversity words(1.0), more focused(0.5) max_output_tokens: long response(500), short & precise(50) top_k: diverse(40), accurate(5)

   Use Case Best Settings Coding assistant temperature=0.2, top_p=0.8, top_k=10, max_output_tokens=300 Factual Q&A temperature=0.1, top_p=0.9, top_k=5, max_output_tokens=200 Casual Chatbot temperature=0.7, top_p=0.9, top_k=40, max_output_tokens=150 Story Writing temperature=1.0, top_p=1.0, top_k=50, max_output_tokens=500

4. Gemini Models & Best Use Cases (with Example)

   1. Gemini 1.5 Pro – Best for complex reasoning, coding, and long-context understanding. Example: An AI assistant that summarizes long research papers.
   2. Gemini 1.5 Flash – Optimized for speed, ideal for chatbots and real-time applications. Example: A customer support chatbot responding instantly.
   3. Gemini 2.0 Flash – Faster than 1.5 Flash, supports multimodal inputs (text, images, audio). Example: A chatbot that processes images and provides instant feedback.
   4. Gemini 2.5 Pro (Experimental) – Most advanced, excels in coding, reasoning, and largescale AI tasks. Example: An AI that debugs and optimizes entire codebases automatically.
      FOR HACKATHONS ( Gemini 1.5 Flash, Gemini 2.0 Flash) For speed → Flash models (e.g., quick chatbot).
      For deep understanding → Pro models (e.g., research summarization).

5. Different Types of Prompting

6. Zero-Shot Prompting – No examples given, the model generates a response based on its training. Example: "Explain machine learning."

7. One-Shot Prompting – One example is given to guide the model’s response. Example: "A cat is a small, furry animal. Now, describe a dog."

8. Few-Shot Prompting – Multiple examples are given to set the response pattern. Example: "A cat is a small, furry pet. A parrot is a colorful bird. Now, describe a dolphin."

9. Chain-of-Thought (CoT) Prompting – The model is guided to reason step-by-step. Example: "Solve: 23 × 7. First, multiply 20 × 7, then 3 × 7, and add both results."

10. Instruction-Based Prompting – Clear instructions improve output quality. Example: "Summarize this paragraph in 3 bullet points."

11. Role-Based Prompting – Assigning the model a role for better responses. Example: "You are a history professor. Explain World War II in simple terms."
    For hackathons, Few-Shot and Chain-of-Thought Prompting work best for accuracy and reasoning.

    FOR HISTORY SAVING : import google.generativeai as genai from flask import Flask, request, jsonify from flask_cors import CORS

Configure API Key

genai.configure(api_key="YOUR_API_KEY") model = genai.GenerativeModel("gemini-1.5-flash")

Initialize Flask app

app = Flask(name) CORS(app)

Dictionary to store chat history

chat_history = {}

@app.route('/chat', methods=['POST']) def chat(): """Handles chat requests and stores history.""" try: data = request.json user_id = data.get("user_id", "default_user") # Unique user ID user_message = data.get("message", "").strip() if not user_message: return jsonify({"error": "No message provided"}), 400 # Initialize history if user is new if user_id not in chat_history: chat_history[user_id] = [] # Add user message to history chat_history[user_id].append({"role": "user", "message": user_message}) # Generate AI response response = model.generate_content(user_message) bot_response = response.text.strip() # Add bot response to history chat_history[user_id].append({"role": "bot", "message": bot_response}) return jsonify({"response": bot_response, "history": chat_history[user_id]}) except Exception as e: return jsonify({"error": "Something went wrong", "details": str(e)}), 500

if name == 'main': app.run(debug=True)