How to Actually Use LLMs in Your Daily Life

This is Part 2 of the AI Agents series. Part 1 covered how LLMs actually work under the hood. This one is practical — here’s how to put them to work.

If you’re already using LLMs every day, you might know most of this. But if you’re a student who’s been hesitant to start, or a developer who only uses them occasionally, there’s probably something here worth picking up.

1. Clear your doubts instantly

The old loop: watch a video → get confused → post in a forum → wait two days for someone to reply who’s also confused.

The new loop: ask Claude, ChatGPT, or Gemini directly.

Example: A video says “your model shouldn’t overfit” but never explains why. Instead of sitting on that question, paste it in:

“What actually happens if a model overfits? Explain it clearly.”

You’ll get a real answer. If it’s still fuzzy, ask for more detail. Keep drilling until it clicks.

95% of doubts you have as a student can be resolved this way — faster and more reliably than waiting for a reply from someone who’s busy with their own life. LLMs aren’t a perfect source of truth, but for conceptual questions at the learning stage, they’re genuinely excellent.

2. Build an ATS-friendly resume

Most resumes get filtered out before a human ever sees them. Applicant Tracking Systems (ATS) scan for keyword matches against the job description — if your resume doesn’t match, it’s gone.

Here’s the workflow:

1. Paste your current resume into the LLM (mask your email and phone if you care about privacy)
2. Paste the job description you’re targeting
3. Prompt: “I have X years of experience. Rewrite my resume to be ATS-friendly for this job description.”

The LLM will restructure your bullets to mirror the language in the JD. This alone meaningfully improves your response rate without you writing a different resume for every job from scratch.

3. Find project ideas (and execute them)

Good projects on your resume open doors. The hard part early on is knowing what to build.

Just ask:

“I’m a 4th-year B.Tech student. Give me ML project ideas. I’m interested in classification and time series.”

You’ll get a list. One common suggestion is stock price prediction — skip that one. Time series models work by finding patterns in historical data, but stock prices are driven by unpredictable daily dynamics that break those patterns. You’ll produce a project that looks impressive but doesn’t actually work well, which is worse than a simpler project done correctly.

Better picks from the same prompt:

• Customer Churn Prediction — telecom, e-commerce, or retail churn datasets. Real business problem, clean structure, interpretable outputs.
• Recommendation System — choose a domain you care about (healthcare, sports, retail). Well-understood problem with a lot of public data.

Once you pick an idea, use the LLM for step-by-step execution guidance — data selection, feature engineering, model training, hyperparameter tuning, and deploying with Streamlit. You don’t need a paid course. The tools are enough.

4. How a working Data Scientist actually uses them

Here’s how LLMs fit into real day-to-day work:

Writing boilerplate code faster Docstrings, type hints, argument validation — tedious to write, easy to generate. Offload it.

Code optimization with GitHub Copilot Rather than manually reviewing your own code for inefficiencies, ask Copilot: “How and where can I improve this code to make it more efficient?” You fix the code and learn something about the pattern you missed.

Brainstorming ML approaches When solving a new ML use case, use the LLM as a sounding board before committing to an approach. Something like:

“If I approach this with method X, where will I hit problems? Is there a better path?”

The key: don’t follow LLM suggestions blindly. Use your own judgment to evaluate whether an idea actually makes sense for your specific constraints. LLMs are great at generating options; you’re responsible for filtering them.

5. Automate repetitive content tasks

A real example from using LLMs for content creation:

The goal was to cut YouTube videos into short Instagram reels without manually scrubbing through footage. The solution:

1. Generate a transcription (subtitles with timestamps) for the YouTube video
2. Paste the timestamped transcript into Claude or ChatGPT
3. Prompt: “Identify the best segments, max 90 seconds each, that would work as standalone Instagram reels”

The LLM returns timestamps. You cut on those timestamps. No manual editing.

The same workflow works for: generating script outlines, drafting video topics, and writing first-draft descriptions for any long-form content.

One critical warning: when not to use LLMs

Don’t reach for the LLM the second you hit friction.

When you’re learning a new concept, the struggle is part of the process. Sit with a problem for at least 30 to 60 minutes before asking. Use LLMs to resolve a doubt after you’ve genuinely tried to work through it — not as an instant escape from thinking.

If you use them as a simple question bank with no effort on your part, two things happen:

• You stop developing your own intuition
• You become dependent on a tool that sometimes confidently gives you wrong answers

LLMs work best as a multiplier on your existing thinking, not a replacement for it.

What’s next

The series continues with:

• Prompt engineering — how to get consistently good outputs
• RAG and vector databases — giving LLMs access to your own documents
• Building real AI agents — tools, memory, and putting it all together

Full video walkthrough is embedded above.