How to Build an App Like Indeed

Aggregation is the cold-start answer

Rather than persuading employers to post, pull postings from company career pages, ATS feeds, and partner boards. Candidates arrive because the coverage is good; employers pay later for placement once you have the audience.

Sourcing paths, roughly in order of reliability:

ATS integrations and feeds — Greenhouse, Lever, Workday and others expose structured job feeds. Cleanest data.
Partner and syndication feeds — standard formats, contractual, reliable.
Crawling career pages — broadest coverage and most fragile; check terms of use for each source.

Deduplication is the real engineering problem

The same role appears from the company site, two aggregators, and a recruiter — with different titles, formatting, and sometimes different salary claims.

Naive matching on title plus company fails immediately: "Senior Software Engineer", "Sr. SWE", and "Software Engineer II" may be one posting or three roles. What works is a composite signal — normalised company identity, embedding similarity on the description, location proximity, and posting date — with a canonical record chosen by source reliability and everything else linked to it.

Getting this wrong is immediately visible to users, and duplicate-choked results are the top complaint about aggregators.

Relevance beats filtering

Job titles are wildly inconsistent across companies. Pure keyword search fails because a candidate searching "product manager" should see "PM", "Product Owner", and sometimes "Program Manager".

Hybrid retrieval — embeddings for semantic similarity plus keyword matching for exact terms like technologies and certifications — handles this far better than either alone. Rank on relevance, recency, and posting quality, and always surface why a result matched.

The apply-flow decision shapes the product

Redirect to the employer — simple, no application data, and you lose the candidate at the handoff.

Apply on your platform — better experience and you own the data, but you must deliver applications into the employer's ATS, which means integration work per system.

Most aggregators do both, using on-platform apply as the premium product. Decide early; it determines your data model and your revenue model together.

Freshness is a trust signal

Expired postings are the fastest way to lose candidates. Re-crawl on a cadence weighted by source volatility, expire aggressively when a posting disappears from source, and show the posting date honestly.

Related builds

How to build a job board — the direct-posting model
AI MVP for HR
AI workflow automation for HR and recruitment

FAQ

How do job aggregators solve the cold-start problem?

By not waiting for employers. They ingest postings from ATS feeds, syndication partners, and company career pages so candidates have a reason to visit, then monetise employer placement once the audience exists. That shifts the hard problems from sales to ingestion, deduplication, and relevance.

How do you deduplicate job postings across sources?

With a composite signal rather than title matching: normalised company identity, embedding similarity on the description, location proximity, and posting date. Title matching alone fails because 'Senior Software Engineer', 'Sr. SWE', and 'Software Engineer II' may be one role or three — and duplicate-choked results are the top user complaint about aggregators.

Should candidates apply on the aggregator or the employer site?

Most run both, with on-platform apply as the premium product. Redirecting is simple but loses the candidate and the data at handoff; applying on-platform gives a better experience and richer data but requires delivering applications into each employer's ATS. The choice determines your data model and revenue model together, so decide early.