What DueCare does, for whom

Six ways teams put DueCare to work.

DueCare runs the same safety-harness pattern for every audience. What changes is the channel, the questions, and what the system hands back. Each section below covers what it does, who benefits, and includes a deployment architecture you can expand inline.

01 · Platform safety screening

Catch exploitative user-generated content.

A pre-publish screening layer for platform safety teams, recruitment marketplaces, and other UGC review surfaces. It reads posts, listings, messages, and ads against corridor-specific rules and returns a flag, a cited reason, and a suggested edit the trust & safety team can act on.

What it does

Capabilities

  • Detects exploitative UGC: illegal recruitment fees, passport retention, contract substitution, deceptive job ads, sham agencies
  • Identifies recurring patterns across posts, accounts, employers, and corridors
  • Flags borderline content for human review with cited reasoning
  • Risk-scores complex migration-exploitation patterns
  • Suggests redline edits so a post or listing meets corridor rules
  • Returns a versioned pack reference for audit trails

Why it helps

  • Cuts manual T&S review load with explainable, cite-able flags
  • Catches exploitation pre-publish, not post-incident
  • Standardizes enforcement across geographies + languages
  • Audit-ready: every flag tied to a vetted pack version
  • Local model. No user content leaves the platform
▸ Deployment architecture›
Stays local · with platform
Platform infrastructure
  • Listing copy, employer identity
  • Trust & safety decisions, audit logs
  • Takedown / edit-request comms
via anonymizer
May cross via anonymizer
Anonymized signals only
  • Aggregate flag-rate per corridor + rule
  • Pack-version + rule-version fired
  • No employer · no listing · no PII
vetted packs
Public hub returns
Pre-publish verdict
  • Flag with cited rule + section
  • Suggested redline edit
  • Versioned pack reference
Local model · grep rules · suggested-edit tool · anonymizer (outbound) Suggestion only. Platform decides every action
02 · NGO & regulator copilot

Speed up case analysis and corridor inspection.

Runs on a caseworker or inspector's own machine. Surfaces relevant law, drafts replies and complaint forms, and gives regulators a corridor-level view for inspection planning. without case data ever crossing the boundary.

What it does

Capabilities

  • Drafts caseworker replies grounded in cited corridor packs
  • Builds complaint forms from a worker statement
  • Generates triage checklists per case type
  • Produces sector + corridor heatmaps for inspection planning
  • Pack-diff alerts when relevant law changes upstream

Why it helps

  • Cuts time-to-first-draft on every case
  • Aligns drafts to current law without manual lookup
  • Reduces staff burden during seasonal casework spikes
  • Aggregate trends inform inspection without exposing complainants
▸ Deployment architecture›
Stays local · NGO / ministry
Caseworker desktop
  • Case files & worker PII
  • Communications transcripts, field notes
  • Inspection findings, enforcement decisions
opt-in only
May cross at k-anon ≥ 30
Anonymized trends only
  • Corridor-level signal trends
  • Pack-version usage telemetry
  • No individuals · no employers · no cases
vetted packs
Public hub returns
Drafts & trends
  • Caseworker: cited drafts, triage
  • Regulator: read-only heatmap
  • Pack-diff alerts on law changes
Local model · pack inspector · complaint draft builder · trends API Caseworker signs every outgoing artifact
03 · Individual worker / mobile

Local mobile help. Designed for the worker’s device and channel.

A worker/mobile sibling of the harness. Cached corridor packs, in-language answers, and a no-raw-upload boundary to the public hub. The target deployment path supports on-device and offline-capable packaging when the selected model build is available; partner-hosted endpoints can use the same workflow today.

What it does

Capabilities

  • Reads a recruiter / employer message and explains rule fires
  • Cites the relevant corridor-pack section + ILO reference
  • Drafts a question to ask a verified caseworker (never auto-sent)
  • Lists registered, corridor-relevant organizations to contact
  • Available in the worker's primary language

Why it helps

  • Workers get faster guidance on what looks legitimate vs. exploitative
  • Pointers to real caseworkers, not anonymous tip lines
  • Raw chats stay in the worker-controlled or partner-controlled environment, not the public hub
  • Lowers the barrier to seeking help
▸ Deployment architecture›
Stays local · worker-controlled
Worker device or trusted channel
  • Recruiter messages (pasted in)
  • Worker name, phone, ID, contracts
  • Full chat history with DueCare
opt-in per turn
May cross at k-anon ≥ 30
Pattern IDs only
  • Anonymized pattern_id (e.g. fee_request)
  • Corridor + sector buckets only
  • No message text · no contact · no time
vetted packs
Public hub returns
Cited reading + draft
  • Plain-language rule explanation
  • Corridor-pack section + ILO ref
  • Draft question, verified caseworker list
Local or on-device-oriented model · grep rules · pack/qa-np · anonymizer (outbound, opt-in) Draft only. user decides whether and where to share
04 · Researcher

Cite-able research on corridors, trends, and policy.

For researchers, policy analysts, and journalists who study migration corridors, exploitation trends, recruitment-market dynamics, and policy impact. Pin a corridor pack at a hash, query the anonymized signal stream, and run the harness on Gemma 4 — every result is reproducible months later.

What it does

Capabilities

  • Studies corridor risk patterns using version-pinned packs and anonymized aggregate signals
  • Tracks exploitation trends over time. fee inflation, contract substitution, recruiter network shifts
  • Measures policy impact: pack-diffs against signal changes after a regulation lands
  • Runs the harness + rubric on Gemma 4 for reproducible model evaluations
  • Submits prompts & rubric proposals to the public review queue
  • Rejects real PII at the inbound scanner. composite scenarios only

Why it helps

  • Reproducible across labs and time, with stable hashed references
  • Anonymized signals + harness enable corridor and policy research without raw case data
  • Comparable results as packs, rules, and rubrics evolve
  • Removes ambiguity from policy and corridor analysis
  • Lowers cost of entry. no expensive infrastructure or data deals
▸ Deployment architecture›
Stays local · institution
Lab workstation
  • Subject data, IRB-controlled material
  • Institutional records & correspondence
  • Working notebooks & intermediate runs
via inbound scanner
Boundary · scrubbed
Public sources only
  • Versioned evaluation packs + rubrics
  • Public-source corridor packs (scrubbed)
  • Submitted proposals → review queue
cite-able hash
Public hub returns
Cite-able results
  • Pack hash for reproducibility
  • Composite-only eval prompts
  • Runnable on Kaggle / GitHub / local
Local model · eval harness · rubric runner · pack manifest Cite the hash. proposals enter the public review queue
05 · Anonymized knowledge sharing

Share verified facts, not files.

Reviewed cases stay inside the local workbench. Reviewers choose candidate facts, local sanitization removes PII and verifies k-anonymity and re-identification risk, and only sanitized knowledge objects reach the public hub. The worker is never exposed; the hub gets stronger anyway.

What it does

Capabilities

  • Fact selection from confirmed evidence graphs
  • PII redaction + k-anonymity check + re-identification risk score
  • Knowledge-object schema with corridor, indicator, evidence row, pii_status
  • Reviewer approval gate before any export leaves the workbench
  • Versioned knowledge packs distributed back to every other client

Why it helps

  • NGOs and regulators contribute without exposing raw case files
  • Moderation, mobile guidance, and research all sharpen as facts accumulate
  • Privacy-by-default. nothing leaves local by accident
  • Every shared fact is auditable back to its evidence row
▸ Sharing flow›
01 · Local
Reviewed case
  • Confirmed evidence graph stays inside the local workbench
  • Raw files never proposed for sharing
reviewer selects
02 · Sanitization check
Sanitization
  • PII removed from candidate facts
  • k-anonymity verified, re-identification risk scored
  • Gate must pass before egress
approved facts
03 · Hub
Knowledge objects
  • Published as versioned knowledge objects
  • Strengthens packs and rules for every client
  • Diff-notifications when packs update
Local reviewed graph · sanitization check · public hub Worker never exposed. Hub gets stronger anyway.
06 · Developer / integration partner

Drop the runtime into your own product or channel.

DueCare ships as a containerized runtime with a small, well-defined API. Wire it into a Messenger or WhatsApp adapter, a moderator console, an internal case-management system, or a sibling app. you own the channel, we provide the harness ecosystem, packs, and model layer.

What it does

Capabilities

  • Containerized runtime image with the full DueCare harness ecosystem
  • Stable HTTP API for screen / explain / draft / cite endpoints
  • Pack catalog you pin per environment
  • Streaming responses + per-turn telemetry hooks
  • Optional bring-your-own model layer (vetted open-weights)

Why it helps

  • Use DueCare inside the channel your audience already uses
  • Keep your existing data model and access controls
  • On-prem or private-cloud; no traffic to a vendor SaaS
  • Same vetted packs as every other deployment
▸ Deployment architecture›
Your environment
Partner infrastructure
  • User identity & channel state
  • Product database & access controls
  • Compliance / audit logs
HTTP API
DueCare runtime container
Self-contained
  • Local model + harness ecosystem
  • Pinned pack catalog
  • Outbound anonymizer at egress
vetted packs
Public hub
Pack distribution
  • Pinned pack downloads
  • Pack-diff notifications on update
  • Optional aggregate telemetry
Container image · HTTP API · channel adapters · pinned packs You own the channel. we provide the harness ecosystem

Need a use case that isn't on this page?

The runtime is open-source and the API is small. If you're working on something corridor-adjacent and want to talk it through, the contribute page has the intake.

Contribute a corridor → Watch the demo