Hello AI Enthusiast,

What if instead of scrolling through endless job postings every morning, you just opened your inbox to find the three or four roles that actually match your profile — already scored and explained?

That's exactly what Matteo Bacchin built during our AI Agent Bootcamp. His project, Navigo, automatically scrapes LinkedIn job listings, compares each one to his CV, and delivers only the strongest matches to his email. No scrolling. No guesswork. Just signal.

Job hunting is a grind. Keeping an eye on the market means checking LinkedIn daily, reading through a dozen descriptions, and realizing most aren't even close to what you're looking for. Too junior. Wrong language requirement. Off-industry. LinkedIn's own recommendations don't help much either — they're optimized for engagement, not genuine fit.

Matteo decided to automate his way out of it.

Navigo connects three tools — Apify, OpenAI, and Make — into a single automated flow that runs daily. Here's how each piece fits together.

The automation starts with an Apify actor — a pre-built web scraping tool — configured to search LinkedIn based on Matteo's desired job titles and locations. Every day, it pulls a fresh batch of job listings automatically. No manual search needed.

This is the part most people wouldn't know how to build — but inside Make, Apify connects like any other module, with no coding required.

Once the listings come in, each job description gets sent individually to OpenAI's GPT-4o model. The AI acts as a senior HR advisor evaluating whether the role is a genuine fit.

For every listing, the model evaluates four dimensions:

Each dimension gets a score from 1 to 10. The model then calculates an overall average and makes a call: Strong Match (score of 8 or above) or No Match.

Matteo ran a comparison between GPT-4o and GPT-4o-mini to decide which model to use. GPT-4o-mini was 16x cheaper, but its accuracy dropped to 70% — it was too loose with its recommendations, flagging jobs that didn't actually meet the threshold. GPT-4o hit 100% accuracy and consistent formatting. For a tool built to filter noise, accuracy mattered more than cost.

At the end of the flow, Make sends a single daily email containing only the Strong Match listings. Each job appears in a clean table with its scores, a short explanation of why it's a fit, and tips on what to highlight in the application.

No fluff. No noise. Just the jobs worth looking at — with the reasoning already done.

Scoping down is a skill. Matteo's original vision was a full product where anyone could upload their CV and set their search terms. Getting a simpler version working came first.

Model choice has real consequences. The model you pick affects both output quality and cost. Testing options before committing saves time and money.

Cost awareness from day one. Apify's LinkedIn scraper costs around $0.25 per run, roughly $7.50/month. Matteo kept AI costs down by embedding a condensed CV summary in the system prompt rather than pasting the full document each time.

You don't need to be job hunting to find this useful. The underlying pattern — scrape data, evaluate it against a personal profile with AI, surface only the best results — applies to a lot of problems:

Start by thinking about something you currently filter manually. Then ask: what's the source of the raw data, what criteria would define a "strong match" for you, and where would you want the results delivered?

That's the whole architecture. Matteo built it in a few weeks, with coaching support, and no prior experience with any of these tools.

Want to get even more practical? Explore hands-on AI learning with AI Academy:

We'll be back with more AI tips soon!