Information, against the brief
Compare the title, date, venue, and contact instructions with the approved source.
Wrong dates and missing details slip through poster approval every day. RayDev uses Claude to read the design and its approved brief side by side, then points reviewers to the exact evidence.
Interactive prototype with prewritten findings. Live Claude analysis is our next milestone.
13:00 — 17:00
STUDIO 04
24 Oct on the poster. 25 Oct in the brief.
A poster can look ready while one critical detail is still wrong. RayDev’s proposed workflow brings the source, the design, and the reviewer together.
Compare the title, date, venue, and contact instructions with the approved source.
Surface readability, placement, and missing elements for a person to inspect.
Keep each concern connected to a poster region and a source requirement. People make the final decision.
Select a concern to highlight its location and compare the evidence. Switch to the revised poster to see the corresponding example edit.
The outlined area is the selected evidence region. You can also select a region directly.
View the selected findingUse the approved details below on the final poster.
Synthetic brief. All names, locations, and event details are fictional.
Local demo state only. This does not approve the poster.
Return to the posterEach sample finding carries a category, source requirement, observed content, and normalized poster region. A review flag records only your choices in this browser session.
The intended MCP-based workflow connects inputs, model-assisted checks, and evidence. It keeps final approval with the team reviewing the work.
Bring the design and its source requirements together.
Claude reads both inputs and checks information and visual concerns.
Return structured observations tied to the source and poster region.
Inspect the evidence, resolve concerns, and decide what is ready.
Poster review is a multimodal comparison: an image on one side, a written brief on the other, and a structured answer people can act on. Each step maps to a Claude capability.
Extract dates, venues, and small print directly from the artwork, and judge readability in context.
Load the approved brief, including PDFs and revision notes, as the source of truth for every check.
Each finding follows the JSON schema shown in the demo: category, source field, observed text, poster region.
Connect to the storage and design tools teams already use, so review happens without re-uploading.
We are a five-person team in an AI education program, building an MCP-based review system for event posters.
The review experience, evidence linking, and finding schema are designed and testable on this page.
Send the poster image and brief to the Claude API and return findings in the same schema, replacing the prewritten examples.
Score detection on a set of synthetic posters with seeded errors, then pilot with event and design teams.
Not yet. The findings shown are prewritten for a fictional event so the review experience can be tested first. Connecting the Claude API is the next milestone.
Not in this version, which uses only the included synthetic samples. Uploads will arrive with live analysis.
No. RayDev surfaces evidence; a person makes the final decision. It does not make legal infringement determinations.