What we've shipped. What's next. Honest dates.
We commit to under-promising and over-shipping. The roadmap below is the current state of plans, with quarter-level granularity where we're confident and no dates where we aren't.
Live now
In progress (Q3)
Active development; shipping within the quarter.
Embedding population at scale
Background pipeline embeds all candidates + jobs + skills with text-embedding-3-large. Activates the embedding-similarity sub-model in the ensemble.
Logistic regression scoring model
First trained sub-model — supervised on observed placement outcomes. Calibrated probabilities for the ensemble weighted average.
Retention model v2 (XGBoost)
Replaces the v1 heuristic. Trains weekly on cumulative retention data. Outputs feed the match score and the retention forecast.
Next (Q4)
Planned for the next quarter. Specs locked.
Internal mobility
Match employer's existing employees to internal openings. Same matching engine, scoped to the employer's own workforce.
Skills assessments v1
Adaptive assessments that verify claimed skills. Generated and graded by Astris HR Model with caseworker validation.
Public API + ATS integrations
REST + webhook surface for matches, placements, and follow-up events. First-class integrations with the top three mid-market ATS systems.
Cross-tenant labor market intelligence
Aggregated, anonymized demand + supply data across the platform. Workforce planners see real-time demand for skills in their region.
Future (no committed dates)
Things we believe in but haven't scoped tightly enough to commit.
Multimodal Astris Model
Fine-tuned model trained on our own retention-labeled corpus. Replaces the LLM reasoning layer over time.
Career navigator (candidate-facing)
Public candidate app: track applications, set preferences, accept matches, message caseworker, view earnings projection.
Employer training marketplace
Connect employers with translation services, onboarding consultancies, and refugee-employment training providers.
SOC 2 Type II + HITRUST
Audit-attested security posture for enterprise and federal-adjacent deployments.