Welcome to Naveen's & Padma's home

Main Content

AI Help

Artificial Intelligence is now useful for learning, writing, summarizing, coding, troubleshooting, planning, image generation, document analysis, and automation. For everyday use, the best approach is to treat AI like a fast assistant: ask clear questions, provide context, verify important answers, and never blindly run generated commands or code.

This page is a high-level reference for AI learning and AI-assisted coding, especially using free plans, local models, open-weight models, and community tools. Free tiers and model limits change often, so always check the official links before depending on a tool for production work.

Where AI is useful

  • Learning: ask for explanations, examples, practice questions, summaries, and step-by-step walkthroughs.
  • Coding: generate boilerplate, explain unfamiliar code, write tests, debug errors, refactor small pieces, and compare design options.
  • DevOps: summarize logs, draft shell scripts, explain Kubernetes errors, build checklists, and review CI/CD failures.
  • Documents: summarize PDFs, rewrite emails, create outlines, and convert rough notes into structured content.
  • Research: collect starting points and compare options, then verify facts with primary sources.

Good AI habits

  • Give context: goal, operating system, programming language, error message, what you already tried, and constraints.
  • Ask for assumptions: List assumptions before giving the answer.
  • Ask for verification steps: How can I test this safely?
  • Use small iterations: ask for one change, test it, then continue.
  • Review code carefully before running it, especially shell commands, database migrations, and cloud changes.
  • Do not paste passwords, API keys, private customer data, medical records, financial data, or sensitive company code into public or unknown tools.

Free and low-cost AI coding options

  • ChatGPT Free: useful for general AI help, learning, debugging, and limited Codex/coding access. Official pricing page: ChatGPT pricing.
  • GitHub Copilot Free: useful inside VS Code and GitHub workflows with monthly limits. Official docs: GitHub Copilot plans.
  • Google AI Studio and Gemini API: useful for experimenting with Gemini models and API prototypes; Google lists free-tier rate limits by model. Official pricing: Gemini API pricing.
  • VS Code AI agents: VS Code includes AI agent workflows through GitHub sign-in and supports bringing your own API key. Official docs: VS Code agents.
  • Aider: terminal-based AI pair programming that can work with cloud or local LLMs. Official site: Aider.

Local AI tools

Local tools are useful when you want privacy, offline experimentation, or no per-message API cost. They need enough CPU, memory, and sometimes GPU/Apple Silicon resources. Smaller models are faster and easier to run; larger models can be better but require more hardware.

  • Ollama: a simple way to download and run local models on macOS, Linux, and Windows. Example commands: ollama pull qwen2.5-coder and ollama run qwen2.5-coder. Official download: Ollama download.
  • LM Studio: a desktop app for discovering and running local models with a graphical interface. Official site: LM Studio.
  • Open WebUI: a self-hosted web interface for local and cloud models, commonly used with Ollama. Official docs: Open WebUI docs.
  • LiteLLM: a developer tool that provides one OpenAI-compatible interface across many providers and local Ollama models. Official docs: LiteLLM docs.

Open-weight model resources

  • Hugging Face Models: a large model directory with filters for text generation, GGUF, Ollama, llama.cpp, and other runtimes. Model page: Hugging Face text generation models.
  • Coding-focused models: look for model families such as Qwen Coder, DeepSeek Coder, StarCoder, Code Llama, and other code-tuned models. Check the model card, license, size, context length, and hardware requirements before use.
  • Model size guide: 1B to 3B models are good for quick experiments, 7B to 14B models are a practical local starting point, and 30B+ models usually need strong hardware.
  • License check: open-weight does not always mean unrestricted. Read the license before commercial or public use.

AI coding workflow

  • Start with a clear task: Explain what this function does and identify risky edge cases.
  • Ask for a plan before edits: Give a minimal implementation plan and list files likely to change.
  • Prefer small patches: Change only this function and explain the test cases.
  • Use tests as guardrails: ask AI to add or update tests, then run them yourself.
  • Ask for security review: Review this code for injection, auth, secrets, and unsafe file handling.
  • Ask for rollback steps before infrastructure changes.

Prompt examples

Debugging:
I am on Ubuntu 24.04. This command fails with the error below. Explain the likely causes, give safe diagnostic commands first, and do not suggest destructive changes until the end.

Code review:
Review this code for bugs, security risks, missing tests, and maintainability. List the highest-severity issues first with file and line references.

Learning:
Explain Docker networking to a beginner using one diagram-like explanation, then give five commands I can run locally to observe it.

Local model selection:
I have a Mac with 16GB RAM. Recommend a local coding model size, explain tradeoffs, and give Ollama commands to try it.

Safety checklist

  • Never publish or paste real secrets. Use placeholders like API_KEY_HERE.
  • Do not run AI-generated rm, chmod -R, chown -R, database, firewall, or cloud commands without understanding them.
  • Keep AI-generated code under version control so you can inspect and revert changes.
  • Run linters, tests, and security scanners after AI-generated changes.
  • Verify legal/license requirements before using generated code or open-weight models commercially.

Recommended starting path

  • For quick learning: start with ChatGPT Free or Gemini in AI Studio.
  • For coding inside VS Code: try GitHub Copilot Free or VS Code agents.
  • For local privacy: install Ollama and try a smaller coding model first.
  • For a local ChatGPT-like interface: pair Ollama with Open WebUI.
  • For serious projects: combine AI with Git, tests, code review, and backups.