Unify Platform Vision
Governments and community organizations struggle to launch benefit programs. Manual document review requires armies of staff. Fraud detection systems slow legitimate applicants. Language barriers exclude entire communities. Setup takes months instead of days. As a result, $140 billion in public benefits go unclaimed every year.Unify is the easiest and best way to design, launch, and scale benefit programs. We’ve processed nearly 500,000 applications across 16 states, proving that speed and dignity can coexist at scale.
The Platform
Unify is not a single application—it’s an integrated platform of specialized tools that work together:Terra
Form builder and intake engine. Create complex, multi-language forms with conditional logic. Collect submissions and sync to case management systems.
Pathfinder
Benefits discovery portal. Help people find programs they qualify for through guided screening and personalized matching.
Forge
Program design studio. Build program templates, generate eligibility criteria, and create all program messaging—powered by AI.
Sentinel
Fraud analysis platform. Define fraud vectors, run batch analysis, and protect program integrity while maintaining low false-positive rates.
Hub
Unified applicant view. See the complete lifecycle of every person across all programs, applications, and benefits received.
How They Connect
Data Flow
- Program Manager uses Forge to design a new benefit program, generating eligibility rules, required documents, and all messaging variants
- Forge exports the program template to Terra as a draft form
- Applicants discover the program through Pathfinder and submit applications via Terra
- Submissions flow to Airtable for case management review
- Sentinel analyzes submissions for fraud patterns, flagging suspicious applications
- Hub shows the complete applicant lifecycle across all touchpoints
Shared Infrastructure
All Unify applications share common infrastructure:| Component | Technology | Purpose |
|---|---|---|
| Database | Supabase (PostgreSQL) | Single source of truth for all data |
| Auth | WorkOS | Enterprise SSO, role-based permissions |
| Identity | @unify/identity | Unified applicant profiles across apps |
| Files | Supabase Storage + Google Drive | Secure document storage |
| Notifications | Resend (email) + Twilio (SMS) | Multi-channel communications |
| Secrets | Doppler | Environment management across apps |
Shared Packages
AI Capabilities
The platform leverages AI to remove barriers at every stage:Document Extraction
Document Extraction
Extract structured data from paystubs, IDs, bank statements, and proof of address. Feed extracted data into Terra forms, fraud analysis, and case management systems.
Fraud Detection
Fraud Detection
Analyze patterns across 450,000+ applications. Identify synthetic identities, coordinated fraud attempts, and document forgery. Maintain low false-positive rates to protect legitimate applicants.
Program Template Generation
Program Template Generation
Analyze existing programs and public benefit models to generate eligibility criteria, documentation requirements, and workflow designs. Compress program launch from months to days.
Multilingual Messaging
Multilingual Messaging
Generate all program communications—social media, notifications, approval letters, outreach materials—and auto-translate to 20+ languages while preserving cultural context.
Personalized Outreach
Personalized Outreach
Inform applicants about relevant opportunities, upcoming deadlines, and next steps in their preferred language and communication channel.
Grant Context
This platform expansion is funded by the GitLab Foundation’s AI for Economic Opportunity grant. The grant supports deploying AI to:- Foundation Phase: Document extraction pipeline, baseline fraud detection, multilingual translation infrastructure
- Intelligence Layer Phase: Cross-program insights, program template generation, personalized outreach engine
- Integration & Validation Phase: Partner validation, measurement, documentation for scaling
Success Metrics
| Metric | Target |
|---|---|
| Document extraction | Reduce manual review from hours to minutes per application |
| Fraud detection | Maintain <15% false positive rate while catching coordinated fraud |
| Language coverage | Expand from 2 languages to 20+ |
| Program launch speed | Compress setup from months to <2 weeks |
| Geographic expansion | Grow from 16 states to 25 states |
| Distribution capacity | Scale from 15M annually |
Architecture Principles
Monorepo, Multiple Apps
Monorepo, Multiple Apps
All applications live in a single repository. Shared packages prevent duplication. Database migrations stay in sync. One team can work across all apps without context-switching.
Human in the Loop
Human in the Loop
Every high-stakes decision (fraud flagging, eligibility determination, benefit denial) requires human review. AI serves as a support tool, not a replacement for human judgment.
Privacy by Design
Privacy by Design
PII is encrypted at rest. Separate queryable metadata from sensitive values. Audit all access. Never log credentials or sensitive URLs.
Applicant Data Is Sacred
Applicant Data Is Sacred
Benefit applicants trust us with their most sensitive information. Every architectural decision prioritizes data protection. Files are never publicly accessible. Submissions succeed even if integrations fail.
Path to Redemption
Path to Redemption
Someone flagged for fraud in one program should not be barred from all future programs without recourse. Fraud flags are program-specific unless confirmed through human review. Clear processes exist to exit fraud status.