Verra
Make yourself comfortable

Make Verra work for you.

Choose what helps you. Calm mode combines comfortable text, less motion and a quiet layout. It is available to everyone, including people who prefer a lower-sensory experience.

Listen to this page

Read the visible visit details aloud. Only on-device voices are used; reading never starts automatically.

Read-aloud is unavailable in this browser. You can still use your device’s screen reader.

Requirements, conflicts, evidence and sharing permissions stay visible and unchanged.

Explore the demo

For a thoughtful first look

Follow an arrangement.
Then test its boundaries.

Explore the problem Verra is built around, the decisions it keeps with the person, and the parts that work today.

Open the judge demo

A route through the demo

  1. Take the short welcome tour, or skip it and open the studio visit’s Evidence. The sample website covers the entrance, while a fictional reply puts the class upstairs with a broken lift.
  2. Review the proposed ground-floor room change. The step-free requirement stays unchanged.
  3. Approve the question, then choose “Load example reply” to advance the demo conversation.
  4. Open the visit plan. Route and seating have supporting sample statements. Quieter arrival remains unknown.
  5. Try pause, cancellation, declining an alternative and resetting the demo. Add a fictional visit or edit the sample access profile.
Start with the evidence

Explore at your pace

Replay the tour from the demo toolbar. Ask the demo companion about evidence, permission or the next step. Its answers use the selected example’s recorded details.

Accessibility offers on-device read-aloud, colour-independent status cues, stronger contrast and calmer reading. Movement follows your device’s reduced-motion setting.

What to look for

Requirement-level clarity, scoped sharing, explicit uncertainty, durable control and useful decisions are central to Verra. The interface should not turn a vague venue statement into a universal accessibility claim.

Try dinner and gallery scenarios too. Access logistics extend beyond events, and previous information does not automatically answer a new visit’s questions.

Implementation details

Implementation status

Working foundation.
Full product in progress.

Implemented and locally verified: personal visit and needs storage, account isolation, permission recording, transactional work requests, safe retries, pause/resume/cancel, personal data export, guarded research jobs, source-backed findings and unsent inquiry drafts. Controlled tests use labelled synthetic model/page inputs. A separate local check used real Luna and a public venue page, saved findings and preserved an unanswered room-route requirement.

Interactive simulation: the judge workspace’s fictional evidence, correspondence, decisions and sample plans. Demo changes persist only in this browser.

Deployed and running: the AgentCore runtime and the scheduled Lambda dispatcher. Not connected: venue email send and receive, follow-ups, and a linked calendar account, though a visit downloads as a calendar file. Hosted sign-in email is rate limited on a shared sender, which is why the demo needs no account. No live efficacy claims are made.

Technical direction

Next.js and Supabase provide the personal application and database boundaries. The research agent is Python Strands running on Amazon Bedrock AgentCore, using GPT-5.6 Luna through the direct OpenAI provider. The AgentCore runtime is deployed and ready, and an EventBridge schedule invokes a Lambda dispatcher once a minute to claim due work. Real model research has been verified against live venue pages. A full hosted run, from a signed-in account through to a saved report, has not been completed yet.

Queued work is accepted only after a database transaction. Future external-action tools must check permission, cancellation and duplicate delivery at execution time.

The full roadmap includes evidence, replies, alternatives, confirmed plans, return visits, companions, notifications, data control and recovery.