Skills-based matching, built for real complexity.
Eight tightly-integrated capabilities. One model. Multiplicative effects.
A knowledge graph mapping skills, occupations, and the bridges between them.
Astris HR maintains a curated graph of skills with bidirectional transferable links. When a candidate is missing a required skill, we look at the skills they do have and check the graph for equivalents. Partial credit is granted, with the bridge logged on the match.
Where curated edges run out, embedding similarity covers the rest. Every claimed skill gets compared semantically — not just exact-matched.
Foreign occupations and credentials are preserved in their source language. The graph maps them into US-anchored equivalents without erasing them.
See the candidate who'd be missed by a keyword filter.
An Agricultural Supervisor with seven years of team leadership and inventory handling is a perfect Warehouse Lead candidate. But no keyword search will find them — the words on their resume do not match the words on the JD.
Astris HR Model identifies the bridges automatically. Crop cultivation transfers to inventory handling. Tractor operation transfers to forklift. Team leadership is team leadership.
Every transferable hit is shown to the employer with the bridge: 'Crop Cultivation → Inventory Handling.' No hand-waving. Every credit is auditable.
Adjacent occupations the candidate could realistically grow into.
Astris HR Model recommends 3-5 career pathways from a candidate's current profile — adjacent roles their existing skills already transfer to, with realistic estimates of how many months of bridging training would unlock each one.
Each pathway carries a confidence score and a list of specific bridging skills + certifications to add. For workforce development organizations, these become the blueprint for upskilling programs.
Seven independent dimensions, one composite score, full transparency.
Skills (40%), experience (20%), language (10%), distance (10%), schedule (10%), certifications (5%), transportation (5%). Each component runs deterministically and explainably. The composite is a weighted average.
Layered on top: company quality boost, transferable-skill credits, and an ensemble framework ready for logistic-regression, gradient-boosted, and neural-net sub-models as training data accumulates.
Every match shows the breakdown, the gaps, the transferable bridges, and a plain-language rationale produced by Astris HR Model.
Probability of 30, 90, 180, and 365-day retention — with the why.
Astris HR doesn't stop at placement. Each match comes with a retention forecast — the probability the candidate is still employed at each milestone. We surface the features driving the forecast: transportation match, childcare match, schedule match, employer training, employer track record.
The forecast feeds back into matching. Candidates with the same skill profile get different recommendations based on which employer has actually retained people like them.
Speak the candidate's language. Always.
Resume parsing in 35+ languages including Pashto, Dari, Tigrinya, Karen, Rohingya, Sorani Kurdish, and Levantine Arabic. UI translated into 8 languages with RTL support.
Follow-up SMS messages go out in the candidate's native language at day 1, 7, 30, 60, 90, 180, and 365. Validation emails. Match notifications. Everything.
Translations get rated by the community. Low ratings auto-trigger Astris HR Model retranslations that override the static copy.
The candidate approves their profile before anyone sees it.
After Astris parses a resume, the candidate gets an email in their native language with a link. They review the parsed profile — skills, languages, work history, education — and either approve or reject with a reason.
Only approved candidates enter matching. Rejections come with a reason that gets routed to the caseworker for correction.
This solves three problems: misparses, misrepresentations, and the bias that comes from employers seeing AI-generated profiles the candidate never agreed to.
The model learns from every rated parse, every employer match feedback, every retention milestone.
Caseworkers, candidates, and employers rate every Astris parse. Low ratings + suggestions trigger automatic re-parse with the hint folded into the prompt.
Translation feedback: thumbs up / down on translated copy. Aggregated low scores trigger Astris-generated overrides that win over the static dictionary.
Employer match ratings: 1-5 stars per match. Becomes training data for the next model retraining cycle.
Placement outcomes: 30/90/180/365-day milestones are the supervised label for retention model retraining.