Skip to content

Feature/model defaults and upload limits - #137

Open
Jordan-Leis wants to merge 5 commits into
mainfrom
feature/model-defaults-and-upload-limits
Open

Feature/model defaults and upload limits#137
Jordan-Leis wants to merge 5 commits into
mainfrom
feature/model-defaults-and-upload-limits

Conversation

@Jordan-Leis

Copy link
Copy Markdown
Collaborator

No description provided.

Jordan-Leis and others added 4 commits July 10, 2026 14:55
Adds gpt-5.4-nano as the top-priority model in the shared
MODEL_PRIORITY ranking, and switches the Study Selection page to
reuse pickBestFromList() instead of its own hardcoded Gemini
2.5 Flash Lite default, so both pages agree on the org default.
…cret

ALLOWED_EMAILS was previously set as two independent plain env vars
(one on the backend container, one on auth-sidecar), which drifted
out of sync in production and locked out a real user. Both containers
now reference one Container-App-level secret via secretRef, so a
single `az containerapp secret set` updates access for both.
The upload size cap was previously three different, inconsistent
hardcoded values (nginx 50MB, backend header-check 25MB, backend
post-read check 20MB), and the frontend claimed a 20MB limit without
enforcing it client-side. Both backend checks now use one env-driven
value (MAX_UPLOAD_SIZE_MB, default 50), matching nginx, and the
frontend enforces + displays the same limit.

Also adds a page-count cap (MAX_PAGES, default 100), enforced at
upload time via PyMuPDF (already a backend dependency) since page
count is otherwise only known after full, slower document processing.
Per feedback from evaluators (via Spencer/boss) — 100 pages was too
restrictive for real studies (300+ page documents already showing
good results), and 500 pairs reasonably with the existing 50MB size
cap (~0.1MB/page). Still fully overridable via the MAX_PAGES env var,
no code changes needed to tune it further.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants