Create Enriched Candidate Lists Using Only Profile Photos

Personalized outreach depends on research most sourcers don't have time to do properly. Checking a candidate's Twitter, Instagram, and Pinterest profiles one by one to find genuine personal detail is slow, and most recruiters end up settling for whatever LinkedIn already shows. An open-source facial recognition tool changes that math by matching a name and photo against multiple social networks automatically and returning a report built for fast, visual double-checking.

The tool in question was built by security researchers as an OSINT utility and works by searching a target's name on a chosen platform, pulling the top search results, downloading their profile photos, and running facial recognition against the original image to confirm a match. It covers Facebook, Instagram, Twitter, LinkedIn, Pinterest, and several international networks, and can run in "fast" mode (stopping at the first likely match) or "accurate" mode (checking further for a better one).

Getting it running is not trivial. It depends on Python, Selenium, GeckoDriver, and a Linux distribution, and installation involves real troubleshooting: choosing between a Windows Subsystem for Linux setup, a rented virtual machine, or a dual-boot install, each with its own networking and dependency headaches.

Once running, the tool takes input as a CSV of names and photo links, a folder of labeled images, or simply a company name, and it can also generate likely email addresses once a company's format is known. Testing against a batch of scraped LinkedIn connections and against a separate list of Slack users compared against another popular scraping tool showed a clear gap: the facial-recognition approach returned fewer total matches but a far higher percentage of correct ones, cutting the manual double-checking time per verified profile roughly in half.

Practical lessons from running it at scale include:

• Using dedicated accounts with real connections and activity improves match rates, since thin or brand-new accounts have less reach into second- and third-degree networks.
• Facebook throttles repeated automated searches quickly, making it unreliable past a handful of requests without account rotation.
• Privacy settings on LinkedIn photos, and mismatched scraping accounts, account for most missed matches rather than tool error.
• Generated email lists still need manual verification, since placeholder or non-employee names can slip through.

Used well, the output turns a slow manual verification step into a side-by-side reference: one screen showing confirmed social profiles, the other for drafting outreach that actually reflects something true about the person being contacted.