Started as an internal verification toolkit for an election-observation deployment. Now an open-source project run by a small team, with a public roadmap, public benchmarks, and a public repo. Apache 2.0 — fork it, run it, contribute back.
Working with verification teams in Kenya, Uganda, and the Philippines, the same pattern kept showing up: a volunteer opens a report, spends ten minutes translating it, looking up the place name, cross-referencing other reports, then thirty seconds making the actual judgement call.
FieldworkIQ does the first ten minutes so the volunteer can spend their day on the thirty seconds. The verifier still reads every report. The verifier still decides what gets published. The map is still your map. We just removed the parts a careful piece of software can do.
Everything we build sits on top of the platforms your team already runs. Ushahidi today. KoboToolbox and ODK Central next. If you've already invested in tooling, FieldworkIQ doesn't ask you to swap it out.
Crisis-mapping runs in vulnerable contexts. The cost of a single mishandled report can be physical safety, deportation, retaliation, or stigma — for the reporter or the people they named. These commitments are architectural, not policy statements. The system enforces them by default.
PII redaction is a workflow-level guardrail, not a post-hoc filter. A public post that still carries a phone number, name, or ID is blocked from publishing — every time.
Graphic violence, sexual violence, child involvement, suicide content — these never bypass human review. Auto-approve is opt-in, low-stakes only, and bounded by the deployment's own threshold.
A single source means "reported," not "verified." We don't inflate confidence to please a dashboard. The verification status reflects the evidence in hand.
Operators can override any classification, reject any drafted post, escalate any case. We record dissents in the audit trail; we never collect metrics that punish high override rates.
No selling, no sharing across tenants, no implicit federation. Reporter identity is stored as a hash, encrypted at rest, with per-tenant key derivation. Right-to-erasure honored within 7 days; 24 hours for safety-critical cases.
They are interpretations grounded in evidence. The verifier approves. The audit trail makes the chain inspectable. The map shows what your team published, not what a model decided.
The full list of ten commitments lives in the repo's ETHICS.md, along with the decision log of how they've evolved. We review this annually with external advisors and disclose materially if we fall short.
You run a deployment, you've read this far, you want to see whether FieldworkIQ fits. A 30-minute call covers your channels, your taxonomy, your verifier team, and whether a two-week trial makes sense.
hello@fieldworkiq.orgThe interactive demo runs in your browser — no signup, no setup. Walk five real cases (a clear publish, a held judgement call, a low-source logistics report) and feel the rhythm of the verifier's workflow.
Open the demo
The repo has Docker Compose, a worked Ushahidi adapter, the eval harness, the adapter SDK. Apache 2.0, no telemetry. Get the dataset access, run docker compose up, evaluate locally.
A pilot is the cheapest way for both sides to find out whether FieldworkIQ helps your team. No hosting fees during the pilot, no commitment past week three. Most teams have decided one way or the other by the end of week two.
We walk your existing workflow. You show us your taxonomy, channels, and verifier load. We sketch what FieldworkIQ would do for you. Either side can stop here.
FieldworkIQ deployed against your Ushahidi instance (managed or self-host, your call). Your verifiers get a one-hour walkthrough. We import your taxonomy and tag rules.
Real reports flow through. Verifiers use FieldworkIQ for their actual queue. We monitor with you — daily check-in on what's working, weekly metrics on time saved and held-case patterns.
Continue (your team takes over, self-hosted or managed); pause (we hand over the data, you keep what was learned); or stop (no obligation either way).
We share a small mailing list of release notes, pilot writeups, and benchmark updates. Roughly monthly. No marketing — just what shipped and what we learned.