Skip to content

Frequently Asked Questions

Quick answers for reviewers, Kaggle users, NGOs, platform teams, researchers, and developers. For exact metrics, use a dated artifact and verification command from Reproducibility.

What This Is

Q: In one sentence, what does DueCare do?

DueCare is an open-source Gemma 4 safety ecosystem that helps local or tenant-controlled deployments review recruitment-risk evidence, draft grounded next steps, and share only reviewed anonymized intelligence.

Q: Is this a chatbot, an app, a library, or a platform?

It is a component ecosystem:

  • Kaggle workbench and live-demo kernels for judges and reviewers.
  • Workbench pages for chat, Bulk File Review, Knowledge Extraction, Search, Templates, and Anonymization & Sharing.
  • Python package surfaces for embedding the harness in other systems.
  • Public hub and GitHub Pages docs for onboarding, APIs, and knowledge-sharing architecture.
  • A-00 and optional benchmark surfaces for reproducible evaluation.

Q: Who is it for?

The public lanes are exactly:

  1. Platform safety
  2. NGO & regulator
  3. Individual worker / mobile
  4. Researcher
  5. Anonymized knowledge sharing
  6. Developer / integration partner

Start with User Paths to choose the right route.

Privacy And Safety

Q: Does my data leave my machine?

Raw worker chats, private case files, IDs, phone numbers, passports, and personal narratives should stay in the local Kaggle session, worker device, trusted NGO machine, or tenant-owned deployment. Sharing is a separate step: only reviewed, redacted fact objects, aggregate signals, pack proposals, benchmark rows, or hash receipts should move to the public hub.

If an operator configures a cloud model or hosted integration, that deployment must document its own data path and consent boundary.

Q: Can the maintainer see my prompts?

Not from a local, Kaggle, or tenant-controlled deployment. The maintainer can only see material that someone submits to a public DueCare service, GitHub, Kaggle, email, or another hosted endpoint. Do not submit raw private case data to the public hub.

Q: Is DueCare certified for SOC 2, HIPAA, GDPR, or FedRAMP?

No. DueCare is open-source software and documentation. A specific deployment may be designed to support an organization's audit or compliance process, but the project itself is not certified.

Q: Is DueCare an emergency service?

No. DueCare can organize information and surface relevant public resources, but emergency support, legal advice, investigation, and enforcement require qualified people and institutions.

Trying It

Q: What is the fastest way to try it?

Use the active Kaggle kernels:

Copy the active kernel.py, use GPU T4 x2 and Internet: On, attach google/gemma-4, then run the notebook. Startup time varies by Kaggle worker, package install path, model cache state, model variant, and Cloudflare quick tunnel availability.

Q: Can it run without a model attached?

Yes for many demo and inspection paths. Heuristic-only mode can serve the start page, slides, deterministic GREP/RAG/tools paths, sample flows, and cached recording slots. Live chat and optional Gemma graph/edge passes need a model.

Q: Can I install it locally?

Use Install, Local Deployment, and the Embedding Guide. Treat the Kaggle kernels as the primary judge-safe path during active judging.

What It Does And Does Not Do

Q: Does DueCare detect trafficking?

It surfaces exploitation and recruitment-risk patterns associated with trafficking. It does not confirm a trafficking case. Confirmation requires investigation, context, and qualified human judgment.

Q: Does DueCare replace lawyers, NGOs, caseworkers, or regulators?

No. It drafts, explains, extracts, checks, and organizes. A trusted person or institution decides what to do.

Q: Does it cover every country or corridor?

No. Corridor and policy coverage is versioned knowledge, not a blanket claim. Use the current corpus manifest or runtime self-audit before naming exact coverage.

Q: Can it read documents and images?

The workbench can organize document bundles and queue/extract text or media evidence depending on the runtime path. Public claims should distinguish deterministic extraction, queued OCR/vision work, and optional Gemma passes.

Research And Claims

Q: Can I cite DueCare?

Yes. Use Researcher Analysis, Reproducibility, and the repository citation metadata. When citing metrics, include the git SHA, dataset/export, model revision, and verification command.

Q: Are the headline lift numbers production guarantees?

No. Treat them as dated smoke/proxy evaluation artifacts unless a current weeks-long or field deployment artifact proves otherwise.

Q: How do I audit public claims?

Run:

python scripts/validate_public_surface.py
python scripts/validate_main_kaggle_kernels.py
python scripts/validate_kaggle_page_sources.py
python -m pytest packages --collect-only -q

Then read Reproducibility for the artifact policy.

Deployment And Integration

Q: I am an NGO or regulator. Where should I start?

Read Caseworker Workflow, NGO Office Deployment, and Regulator Pattern Analysis.

Q: I am a platform team. Where should I start?

Read Enterprise Pilot and Platform / On-Prem Deployment.

Q: I am a developer or integrator. Where should I start?

Read Install, Embedding Guide, and the OpenAPI spec.

Q: I am interested in anonymized knowledge sharing. Where should I start?

Read Information Sharing Architecture, Anonymization Policy, and Submission Labeling Policy.

Other

Q: Is this affiliated with Google?

No. It is an independent open-source project built for the Google Gemma 4 Good Hackathon and uses Gemma 4 as a model layer.

Q: Where are the public URLs?

Q: I have a question that is not here.

File an issue at https://github.com/TaylorAmarelTech/gemma4_comp/issues or email amarel.taylor.s [at] gmail.com with subject [duecare question].