You have a database with ten thousand contacts.

You want to delete the automated bots and keep the real people.

But, If you ask a standard AI chatbot to sort them, it will try to start a conversation.

It will think about the first row of data, write a polite sentence about why it made its choice, and generate a lengthy explanation for a decision that should have taken a fraction of a second.

Then your application has to read that generated paragraph, try to extract the actual answer, and hope the AI did not hallucinate extra details.

By the time it finishes processing fifty rows, you are out of patience and you have spent too much money.

This is the fundamental problem with using large language models for bulk tasks. They think and write one word at a time. When you just need a binary decision, that conversational padding becomes a massive liability.

A new model called Jev is taking a completely different approach.

Built by one of the co-creators of ChatGPT, it strips away the conversational interface entirely. Instead of a chat partner, Jev is an incredibly fast, highly intelligent yes-or-no button.

Now, How to build tools with Jev?

To use Jev effectively, you do not talk to it directly in a chat window. Instead, you use a coding assistant like Claude or ChatGPT to build a lightweight script or web app that sends multiple-choice questions to the Jev API behind the scenes.

Here are three copy-and-paste prompts you can give to your AI coding assistant to build custom, high-speed classification tools for your own workflow.

1. The LinkedIn Network Organizer Prompt

If you have spent years on LinkedIn, your connections list is likely a mess of relevant prospects, recruiters, inactive accounts, and spam. Sorting through tens of thousands of profiles manually is impossible.

You can use this prompt to have Claude build a local web app that filters your exported LinkedIn data using Jev's high-speed decision engine.

Copy and paste this prompt into Claude:

Build a local web app called "LinkedIn Checker" using Next.js and Tailwind. The app should allow me to upload my exported LinkedIn Connections CSV file. It must process the data entirely in the browser, only sending specific fields to the Jev API via the TypeSafe SDK.

Set up Jev to evaluate each contact with these multiple-choice questions:

Role: Is this person an executive, sales professional, or unknown?
Company Type: Does the company name suggest a massive public corporation, a professional services firm, or a family-owned business?
Disqualified: Does the profile belong to a student, job seeker, or recruiter?

Have the app run these questions concurrently for every row, then sort the contacts into priority tiers based on the confidence scores. Give me the code and step-by-step instructions to run it locally.

This prompt forces the AI to build a tool that asks highly specific, restricted questions. By running these targeted queries concurrently, Jev can score thousands of profiles in seconds. You instantly find the few hundred people who actually matter for your next project.

2. The Gmail Contact Cleaner Prompt

Google automatically saves an email address every time you reply to a newsletter, a customer support ticket, or a random inquiry. Over time, your "Other Contacts" list becomes thousands of rows long, filled with useless addresses and forgotten names.

Use this prompt to build a contact sorter that groups your Gmail exports into a deletion pile and a keep pile.

Copy and paste this prompt into Claude:

Build a single-page web app called "Contact Surfer" to help me clean my Google 'Other contacts' CSV export. Use Next.js and the TypeSafe SDK to route classifications to the Jev API.

Before calling Jev, write code to automatically flag common disposable domains or emails with 'no-reply' in the address. For the remaining ambiguous contacts, ask Jev these questions:

Contact Kind: Is this a real individual, a shared role inbox (like support@), an automated sender, or an institutional ID?
Domain Kind: Does the domain look like a personal freemail account or a legitimate corporate organization?

Display the results in a table, grouped into 'Keep', 'Review', and 'Remove' buckets based on Jev's confidence scores. Add a button to download the cleaned CSV so I can re-import it into Google.

Where Jev shines is in the ambiguous cases. Your application can confidently group the contacts, allowing you to instantly delete the newsletters and support desks while preserving genuine human connections.

3. The Research Paper Filter Prompt

For professionals in fast-moving fields like artificial intelligence or medicine, keeping up with daily research is exhausting. Hundreds of new papers might hit archives every day.

You can use this prompt to create a daily filter that reduces hundreds of dense academic abstracts to the three papers that are actually worth your time.

Copy and paste this prompt into Claude:

Write a Python script that downloads the daily metadata for new papers in the cs.AI category on arXiv. I want to filter these papers to find actionable techniques for non-technical users.

Use the TypeSafe SDK to send the title and abstract of each paper to Jev, asking these questions:

Primary Contribution: Is this paper about hardware infrastructure, model fine-tuning, or practical prompting techniques? (Drop anything that requires model training).
Audience Fit: Is this useful for a general knowledge worker, a researcher, or a developer?
Usefulness Score: Score the paper from 1 to 5 on how actionable it is for someone typing into a standard chat interface.

Rank the surviving papers by their usefulness score and output the top results into a clean markdown file with links to the original papers.

This acts as a brutal technical filter. If Jev identifies a paper as hardware infrastructure, it gets dropped immediately. The script only surfaces papers that match your specific workflow.

We are now entering a phase where we stop asking single, massive AI models to do absolutely everything.

Conversational models are incredible for drafting, brainstorming, and writing code.

They are simply the wrong tool for bulk decision-making.

By offloading your classification tasks to a high-speed model like Jev using the prompts above, you save hours of manual sorting.

More importantly, you unlock the ability to finally organize the massive streams of data that define modern digital life.

In case we are meeting for the first time, come over here, it'll be worth the roller coaster of articles that are gonna come up in the next few weeks.

If you're an established writer, here are the brands paying for sponsored articles.