LLM-Based AI Programs

This repository contains a collection of Python AI programs that demonstrate how to build simple interactive applications using Large Language Models (LLMs). The examples include usage with OpenAI models as well as open-source models via OpenRouter. Langchain is used throughout the project.

🚀 Features

• Asynchronous interaction with LLMs using asyncio.
• Support for OpenAI's GPT models (e.g., gpt-4.1-nano).
• Support for free open-source models (e.g., oss120 via OpenRouter).
• A simple CLI loop that:
• Accepts user questions
• Reformulates them into simple prompts (e.g., "Explain like I’m 5")
• Displays model responses interactively

Setup

1. Clone this repository

 git clone https://github.com/yourusername/llm-ai-programs.git
 cd llm-ai-programs

2. Create a virtual environment (optional but recommended)

python -m venv venv
source venv/bin/activate   # Linux / macOS
venv\Scripts\activate      # Windows

3. Install dependencies

pip install openai asyncio

4. Set environment variables

export OPENAI_API_KEY="your_api_key_here"
export OPENROUTER_API_KEY="your_api_key_here"

Learnings :

Model Selection Matters

OpenAI Model Cheat Sheet (Text vs Reasoning)

Model	Type	Strengths	Weaknesses	Best Use Cases	Token Usage Notes
gpt-5-nano	Reasoning	Strong logical reasoning, planning, math, structured problem-solving	Overuses reasoning tokens even for simple tasks (expensive), slower	Multi-step reasoning, coding with planning, proofs, logic-heavy tasks	Consumes reasoning tokens in addition to input/output
gpt-4.1-nano	Non-reasoning	Lightweight, cheap, fast, good for everyday text generation	Not great at deep reasoning or complex logic	Creative writing, chatbots, summarization, Q&A, brainstorming	Counts only input/output tokens, no reasoning tokens
gpt-4.1-mini	Non-reasoning	Balance of speed, quality, and cost; stronger than nano at language tasks	Slightly more costly than nano	General-purpose text tasks, summarization, customer support bots	Input/output tokens only
gpt-4.1	Non-reasoning	High quality, reliable text generation, broad domain knowledge	More expensive and heavier than nano/mini	Professional writing, complex summaries, content generation	Input/output tokens only

---

🔑 Quick Rules of Thumb

• ✅ Use 4.1-nano / 4.1-mini → for creative writing, summaries, general chat, fast responses.
• ✅ Use 4.1 → when you need high-quality outputs and can afford more cost.
• ✅ Use 5-nano → only if your task truly needs deep reasoning or planning.