Preliminary score
19.5/ 100Earlyaverage of 2 judges
- Value for the user8.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI6.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing2.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility2.0 / 15
Data requirements, running costs and a clear next step.
- Originality1.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary16/100
- Strengths
- Your project targets a real clinical need: helping doctors assess brain scans for possible cancer. Your proposed scan-upload-to-assessment flow provides a clear input and intended output.
- To improve
- Your readable materials should include an end-to-end example showing a scan, the AI's output, and how a doctor would use that result. Your testing should identify the dataset, scan modality, and number of held-out labeled cases, with measured results and examples of false positives and false negatives. Your enterprise case needs a specific clinical workflow and an adoption metric, such as review time saved without reducing detection sensitivity. Your feasibility plan should explain data permissions, patient-image privacy, running costs, and a realistic next validation step with clinician oversight.
- Claude Opus 5.5Preliminary23/100
- Strengths
- Your team picked a high-stakes, real problem: helping doctors spot brain tumours on scans, where faster triage could matter. You appear to have shipped a deployed web app with a Supabase backend and recorded a demo video, so there is at least an attempt at an end-to-end flow from image upload to result.
- To improve
- Your submission gives almost no evidence. The README is only a title, and the testing section says only that you tried brain scan images, with no number of cases, accuracy, false positives or negatives, or failure examples. Run a labelled public set such as a Kaggle brain MRI tumour dataset through your app and report sensitivity and specificity, and compare that with a dedicated CNN baseline and with how radiologists triage today. Explain what the model actually does and why a general LLM is trustworthy for medical images, and cover cost per scan, data privacy, clinical validation and a concrete next step such as a pilot with one radiology department.
The team
Abdul Rafay IkramLead
developer
Soufiane Samir
business
Toghrul Aliyev
developer
Karan Shrees
developer