Cerate App

1. 1Target stack: Next.js hosted on Vercel, Neon database, and Gemini Flash 3 for text analysis with cumulative cost tracking.
2. 2Core functionality: PDF text extraction into Markdown, metadata parsing, and relative grading against defined scientific criteria.
3. 3Evaluation strategies: Relative grading calibrated either against the best submission or anchored to the weakest paper as a baseline grade of 4.
4. 4Prompting techniques: Role-based user stories ('as admin i can...'), sequential workflow steps, and a chain-of-thought trigger ('think hard').
5. 5Test data request: Generating 15 mock PDF files based on the current top news story from orf.at.

create an app für tools section.
we need an app to grade papers from students.
host on vercel, use my allready existing accoutn on vercel next js.
use neon database, create new db.
the grading uses the system 1-5 from austria schools, alternatively use 1-6.
for analysing use a good writing model we should use gemini flash 3. show costs for each project (commulated)
each student has the following data. first name, lastname, email. projects have one or more studensts.  if there are just uploads read the name and email from the pdf.



as admin i can create, add, upade and delete projects.
as admin i can create a project with titel, description, grading-notes, default: compare and use worst file from students, second option is orient on best pdf and set notes.
as admin i can see the result of grading (short version) of our critereai. will be generated for each file of students.
as admin i can also see the list of 
think hard and create a list of criteria for grading student papers. important scientific criteria must be included.



there are a few steps to fullfill.
1. read and import pdf files to md. save the data in the database (only update if triggeded by user.)
2. compare all uploaded and converted files. use your defined criteira. 
3. we can either grade the students realtiv to the best work (use criteria) or use the worst work (if not catostrophic the worst work will be graded with 4)

generate 15 pdf files from the actual top thema of orf.at