We measure hiring by who's still there in a year.
Astris HR was built on a simple observation: the labor market for refugees, immigrants, and second-chance workers is full of people who could be excellent employees and full of employers who want them. The failure mode is not lack of supply or lack of demand — it's a matching layer that doesn't understand what either side is actually offering.
Resume-to-JD matching was built for a labor market that no longer exists.
The applicant tracking systems most employers use today were designed for a workforce that wrote standardized English resumes for standardized job descriptions. They optimize for one metric: application volume.
That model fails the people we serve. An Agricultural Supervisor from Kandahar with seven years of team leadership and inventory experience does not show up in a keyword search for "Warehouse Lead." A Tigrinya-speaking nurse aide does not show up in a search that assumes English-only resumes.
And measuring success at placement — instead of at the 90-day mark — hides the failure modes that actually determine whether a household stays out of poverty.
Astris HR Model: matching that understands real candidates and real jobs.
Astris HR Model is our proprietary workforce intelligence model. It reads resumes in 35+ languages, preserves foreign credentials, detects transferable skills through a curated knowledge graph, and scores each candidate-job pair across seven independent dimensions.
Every match comes with a retention forecast and a plain-language rationale. Every match's quality gets rated. The model learns from every real placement on the platform — and the things it learns aren't trapped in a proprietary black box. Caseworkers, candidates, and employers all see the same evidence.
This is the platform we wanted to exist. We're building it because it didn't.
How we make product decisions.
Retention is the unit of value
Every product decision is judged by whether it improves the probability that a placed worker is still employed 90, 180, and 365 days later. Application volume is a vanity metric.
Transparency over polish
Every score is shown with its breakdown. Every rationale is in plain language. Every model decision is auditable. We'd rather be a little ugly and a lot honest.
The candidate consents to everything
Astris parses a resume; the candidate has to approve the parsed profile before any employer sees it. No surprises. No misrepresentation. No assumption that data accuracy is automatic.
Multilingual is not a feature
It's the floor. If a system only works in English, it can't serve the labor market we're built for. We localize first and translate-around second.
Caseworkers are not an afterthought
In refugee and second-chance hiring, the caseworker is the central node. We design for them as a first-class user — their workflow, their portal, their analytics — alongside employers and candidates.
Learn from real outcomes, not synthetic ones
Astris HR Model retrains on observed placements, observed retention, and observed termination reasons. Not on borrowed datasets or scraped résumés.
Who we serve.
Refugee resettlement agencies
Federally-funded resettlement organizations placing newly-arrived refugees into employment as part of the ORR initial reception window.
Workforce development boards
Local and regional workforce agencies serving immigrants, ESL learners, and second-chance workers.
Mission-driven employers
Companies actively building inclusive talent pipelines and willing to invest in onboarding, training, and language support.