Fix SigLIP text padding (max_length) — cross-modal was returning random images

Findings: single load-only fade, no scroll animation

Agent shows the tools it called

Per-session agent memory + new-chat reset

Share uploaded image between Visual QA and OCR Lab

Make findings page fully static (remove scroll animation)

Agent thumbnails bigger + captioned; COCO train2014 fallback

Findings reveal: trigger early + safety net so nothing stays hidden

no-cache header so UI updates show without hard refresh

Agent shows the images it retrieved

Findings: gentle scroll-reveal, drop jumpy bar grow

Findings as a full multi-phase page with examples

Add Findings page; fix OCR-tool prompt + UX hints

Pin transformers 4.57.6 + coerce SigLIP output (fix cross-modal)

Fix COCO image URLs (val2014, not val2017)

Optional password gate (basic auth)

Custom FastAPI website (replaces Gradio UI)

Lazy-import aiohttp/datasets so app imports under lean deps

Add sentencepiece + protobuf for SigLIP tokenizer

Use CPU torch + lean deps to fix Space build

VLPS Gradio demo (Docker Space)

initial commit