import os
import json
import google.generativeai as genai

# NOTE: Apni API key environment variable mein set karna!
# export GOOGLE_API_KEY="YOUR_API_KEY"
try:
    genai.configure(api_key=os.environ["GOOGLE_API_KEY"])
except KeyError:
    print("Bhai, GOOGLE_API_KEY environment variable set karo pehle.")
    # Exit or handle gracefully in a real app
    exit()

MODEL = "gemini-1.5-flash-latest"

def get_initial_filter(market_title: str, recent_news: list[str]) -> dict:
    """
    Tier 1 AI: Use free Gemini model for a quick, cheap first pass.
    """
    # Simple prompt, no fancy stuff
    prompt = f"""
    Market Title: "{market_title}"
    Recent News Headlines: {', '.join(recent_news)}

    Based on the above, is this market worth a deeper look for a short-term trade?
    Consider volatility, news sentiment, and if the topic is currently active.
    Answer in JSON format with two keys: "proceed" (boolean) and "reason" (string, max 15 words).
    Example: {{"proceed": true, "reason": "High volatility and recent positive news."}}
    """
    
    try:
        model = genai.GenerativeModel(MODEL)
        response = model.generate_content(prompt)
        # Basic parsing, real code mein error handling zaroori hai
        result = json.loads(response.text)
        return result
    except Exception as e:
        print(f"Gemini call fail ho gaya: {e}")
        return {"proceed": False, "reason": "API error"}

# --- Example Usage ---
# Yeh data hamara scanner.py laaye ga
market_data = {
    "title": "Will Pakistan win the next T20 match against India?",
    "news": ["Babar Azam scores a century in practice match.", "Shaheen Afridi declared fit."]
}

# Chala ke dekhte hain
initial_assessment = get_initial_filter(market_data["title"], market_data["news"])
print(f"Tier 1 (Gemini) Assessment: {initial_assessment}")

# Output (example):
# Tier 1 (Gemini) Assessment: {'proceed': True, 'reason': 'Positive player news indicates potential for market movement.'}

import os
import json
import anthropic

# NOTE: Apni API key environment variable mein set karna!
# export ANTHROPIC_API_KEY="YOUR_API_KEY"
try:
    client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
except KeyError:
    print("Bhai, ANTHROPIC_API_KEY environment variable set karo pehle.")
    exit()

MODEL = "claude-3-haiku-20240307"

def get_candidate_analysis(market_title: str, price_history: list[float], gemini_reason: str) -> dict:
    """
    Tier 2 AI: Use fast and cheap Haiku for detailed analysis of promising candidates.
    """
    prompt = f"""
    You are a junior trading analyst. Your job is to filter out weak candidates.
    
    Market: "{market_title}"
    Tier 1 Reason to Proceed: "{gemini_reason}"
    Recent Price History (last 24 hours, in cents): {price_history}
    
    Analyze the provided data. Is there a clear, actionable pattern or trend based on our simple momentum strategy?
    - Look for consistent upward or downward movement.
    - Avoid markets that are flat or extremely erratic.
    
    Provide your analysis in a JSON object with three keys:
    1. "is_strong_candidate" (boolean): True only if a clear pattern exists.
    2. "confidence_score" (float): A score from 0.0 to 1.0.
    3. "analysis_summary" (string): A 2-sentence summary of your findings.
    """
    
    try:
        message = client.messages.create(
            model=MODEL,
            max_tokens=200

Multi-tier AI systems allow cost-efficient filtering without sacrificing decision quality.

---

## 📺 Recommended Videos & Resources
- **[Claude API Documentation](https://docs.anthropic.com/)** — Anthropic Claude models and API
  - Type: Official Documentation
  - Link description: Reference for Claude 3 Haiku and other models
- **[Gemini AI Studio](https://aistudio.google.com/)** — Google Gemini free tier access
  - Type: Official Tool
  - Link description: Test Gemini models with no API key required initially
- **[LLM Cost Comparison & Optimization](https://www.youtube.com/results?search_query=llm+api+cost+optimization+claude+gemini)** — Choosing cost-effective models
  - Type: YouTube
  - Link description: Search "LLM cost comparison and optimization"
- **[Prompt Engineering for Trading](https://en.wikipedia.org/wiki/Prompt_engineering)** — Crafting effective prompts
  - Type: Wikipedia
  - Link description: Learn structured prompting for consistent model outputs
- **[Structured JSON Outputs from LLMs](https://www.youtube.com/results?search_query=claude+json+output+parsing)** — Getting parseable model responses
  - Type: YouTube
  - Link description: Search "Claude JSON mode structured outputs"

---

## 🎯 Mini-Challenge
**5-Minute Practical Task:** Write a function that calls both Gemini Flash (free tier) and Claude Haiku to analyze the same market. Compare their responses and confidenceScores. Return which model's analysis to trust more, and calculate the cost difference (Gemini free vs Claude paid per call).

---

## 🖼️ Visual Reference

---