The brief
Hundreds of entries,
a short window to read them.
Jurors read hundreds of entries in a short window. Each one carries long form answers and PDF attachments; reading them one by one, checking the figures and putting a category’s projects side by side takes hours. Until now the informal fix was pasting an entry into a personal AI account, and the entrant’s data went wherever that account went. This project closes that gap: a juror reads cards prepared for their assigned categories and asks questions answered from the entry’s own text. The assistant is a reading aid beside the juror, not a decision maker.
In the jurors’ hands
A card, a table
and a chat that quotes the entry.
Every entry came with an evidence card, every category with a comparison table that shows answers and no totals, and every juror with a chat that answers from the cards. What each of them does, and refuses to do, is set out on the product page.
In use
Wundam Alfred,
in use at The Hammers Awards.
* Screenshots from the synthetic-data demo. No real entries, entrants or jurors.
The season
One season,
start to finish.
Development did not stop at launch: the operations and reporting layer kept growing through the season, while what jurors see and the rules it runs by stayed fixed.
- 711
- entries
- 103
- categories, 66 open to pre-assessment
- 650
- evidence cards open to jurors
- 112
- jurors
- The season ran on it: a pre-assessment window from 11 to 20 September 2026, an in-person jury from 23 to 25 September, the ceremony on 13 October.
Architecture
All on Google Cloud,
all in the EU.
- The whole system runs on Google Cloud: the application on Cloud Run in one EU region, the database on Cloud SQL behind a private IP with no route to the internet, files on Cloud Storage, the model on Vertex AI with Gemini at the EU multi-region endpoint, batch jobs on Cloud Run Jobs, secrets in Secret Manager and jury mail through Google Workspace.
- The application and the model run in a single EU region, enforced in code. The cloud provider has given a written commitment not to train on this data.
- Google Cloud's Data Processing Addendum was accepted on 31 August 2026. The global model endpoint was deliberately not used.
- One provider, one data-processing agreement, one transfer basis. There is no separate vector database, mail service or analytics tool, and none will be added.
Roles and data protection
Clear on paper
before a single entry.
- Wundam LLC acts as data processor. Pazarlamasyon is the data controller.
- Google Cloud is the sub-processor. The programme's application booklet names Wundam LLC and Google Gemini to every entrant.
For awards organisers
Running an awards programme?
Pilot it on one category.
We set it up with your categories, your criteria and your data-protection terms. Your jurors keep scoring where they always have.




