COST OPTIMIZATION:
PURA BUDGET BACHAO, BEST RESULTS PAO.

# Resource monitoring example
import psutil

def check_resource_usage():
 """Current resource usage dekho"""
 cpu = psutil.cpu_percent(interval=1)
 memory = psutil.virtual_memory()
 
 print(f"CPU Usage: {cpu}%")
 print(f"Memory: {memory.used / 1e9:.1f}GB / {memory.total / 1e9:.1f}GB")
 print(f"Memory Usage: {memory.percent}%")
 
 # Right-sizing recommendation
 if memory.percent < 30:
 print(">>> Recommendation: Smaller instance use karo!")
 elif memory.percent > 80:
 print(">>> Warning: Upgrade karo!")
 else:
 print(">>> Size sahi hai")

check_resource_usage()

# Spot vs On-Demand cost comparison
pricing = {
 "g4dn.xlarge": {
 "on_demand": 0.526, # per hour
 "spot": 0.158, # 70% cheaper!
 "monthly_on_demand": 0.526 * 24 * 30,
 "monthly_spot": 0.158 * 24 * 30
 }
}

instance = pricing["g4dn.xlarge"]
savings = instance["monthly_on_demand"] - instance["monthly_spot"]

print(f"On-Demand: ${instance['monthly_on_demand']:.0f}/month")
print(f"Spot: ${instance['monthly_spot']:.0f}/month")
print(f"Savings: ${savings:.0f}/month ({savings/instance['monthly_on_demand']*100:.0f}%)")

# Output:
# On-Demand: $379/month
# Spot: $114/month
# Savings: $265/month (70%)

from functools import lru_cache
import time

@lru_cache(maxsize=100)
def expensive_prediction(text):
 time.sleep(1) # Simulate API call
 return f"Prediction for: {text}"

# First call � full time lagega
start = time.time()
expensive_prediction("hello")
print(f"First: {time.time()-start:.2f}s")

# Cache hit � instant response
start = time.time()
expensive_prediction("hello")
print(f"Cache hit: {time.time()-start:.4f}s")

# Output:
# First: 1.00s
# Cache hit: 0.0001s

# Single vs Batch API costs
cost_comparison = {
 "gpt-4": {"cost_per_1k": 0.03, "quality": "high"},
 "gpt-3.5": {"cost_per_1k": 0.002, "quality": "medium"},
 "llama-2": {"cost_per_1k": 0.001, "quality": "medium"},
}
print(cost_comparison)

# Cost calculation
tokens_per_request = 500
daily_requests = 10000

for model, info in cost_comparison.items():
 daily_cost = (tokens_per_request / 1000) * daily_requests * info["cost_per_1k"]
 monthly_cost = daily_cost * 30
 print(f"{model}: ${monthly_cost:.0f}/month")

# Output:
# gpt-4: $4500/month
# gpt-3.5: $300/month
# llama-2: $150/month (self-hosted)

# Model routing based on complexity
def select_model(query):
 """Query complexity ke basis pe model select karo"""
 
 simple_keywords = ["hello", "thank", "yes", "no", "hi"]
 medium_keywords = ["explain", "summarize", "translate"]
 complex_keywords = ["analyze", "compare", "reason", "code"]
 
 query_lower = query.lower()
 
 if any(word in query_lower for word in complex_keywords):
 return "gpt-4" # Complex tasks
 elif any(word in query_lower for word in medium_keywords):
 return "gpt-3.5" # Medium tasks
 else:
 return "llama-2" # Simple tasks � cheapest!

# Test
queries = [
 "Hello!",
 "Explain quantum computing",
 "Analyze this code and find bugs"
]

for q in queries:
 model = select_model(q)
 print(f"'{q}' -> {model}")

# Output:
# 'Hello!' -> llama-2
# 'Explain quantum computing' -> gpt-3.5
# 'Analyze this code and find bugs' -> gpt-4