Fetching metadata from the HF Docker repository...
Aug 17 08:55 feat: add customer reviews with admin approval and inquiry support system Aug 9 08:14 fix: update landing page and dashboard subtitle copy to UAE regulations 07:42 feat: visitor tracking analytics, landing page & dashboard copy updates, audit purge Jul 28 23:11 Fix extraction deadlock: restrict pdfplumber table extraction to schedule drawings to prevent O(N^3) CPU hangs on dense floor plans 21:13 Fix extraction stopping mid-process: increase gunicorn timeout from 30s default to 600s 19:43 Fix: restore 5 verified Gemini models + fast-fail on invalid model errors instead of hanging 16:31 Fix broken image preview by adding auto-retry for images still being processed in the background 16:23 Restore previous models and combine with new ones 16:21 Update API rotation to use 3 models (2.0-flash, 1.5-flash, 1.5-flash-8b) to improve quotas 15:57 Fix extraction hanging indefinitely by adding fast fail logic for API rate limits and errors 15:34 Add ARCH debug logging to auto_classify 15:29 Fix all pages classified as unknown: re-extract full text from cached PDFs on disk for accurate keyword matching instead of using truncated 250-char DB previews 15:12 Fix 502 Gateway Timeout in Auto Classify. Removed synchronous PDF fallback which blocked HTTP. Added DB error checking to Upload. Truncated texts to 250 chars to prevent MySQL packet size limit errors. 14:58 Fix 502 Gateway Timeout on Saving Classifications by moving AI memory recording to BackgroundTasks 14:50 Combine structural and architectural uploads into a single POST request in Workflow.jsx to cut upload time in half 14:47 Fix 502 Proxy Timeout by moving PDF image rendering to BackgroundTasks. This makes uploads 10x faster and prevents Cloudflare/Nginx from dropping connections. 14:40 Fix OOM crashes on Render proxy by parsing PDFs using mmap from disk instead of massive byte duplication in memory 14:31 Fix global exception status codes and enable origin regex CORS for cross-domain iframes 14:19 Fix 502 payload limit: send str_files and arch_files in separate requests to stay below 10MB proxy payload limit 14:14 Add per-page rendering try-except resilience to prevent PDF parsing crashes during upload 14:11 Fix 502 upload root cause: single FormData POST + top-level try-except error handling + dual-table project ownership check 14:06 Fix 502 upload crash: add proactive garbage collection (gc.collect) to keep RAM under 80MB on cloud containers (Render) 14:03 Fix 502 upload root cause: fix zero-byte file limit cap for free/trial users (PLANS[0]) + add exception handling to upload pipeline 13:57 Optimize render speed to 1200px (0.02s/page) + remove redundant PDF text re-extraction in auto_classify to permanently fix 502 gateway timeouts 13:53 Fix 502 crash root cause: replace PIL Image.frombytes with native C++ PyMuPDF pix.tobytes('jpg', jpg_quality=95) 13:47 Fix classification syntax bug + preserve 4000 char page text + enrich UAE drawing keywords + fall back to PDF text re-extraction 13:39 Fix root cause: lower state limit to 4MB, set max_allowed_packet=64MB on pymysql, add read/write timeouts 13:33 Centralize all state_data writes into upsert_active_state() with auto-trim and size guards - prevents all future save failures 13:29 Fix all SQL dialect errors + truncate state_data + add error logging to save_classification 13:21 Replace dialect SQL queries with universal SELECT -> INSERT/UPDATE logic to prevent database engine syntax errors 13:12 Sanitize classifiedPages payload before posting to /classify/save 13:06 Fix MySQL ON DUPLICATE KEY UPDATE syntax in active projects query 12:59 Eliminate PyMuPDF thread-unsafe access causing process SIGSEGV crash and HTTP 502 error 12:56 Ensure save_file is best-effort to prevent upload failures 12:52 Make auto-classify instant (<0.1s) by bypassing remote vision calls to eliminate 502 proxy timeouts 12:46 Use UPSERT in save_classification to guarantee saving project state without 400 No active project errors 12:37 Expose detailed error message on classification save failure 12:19 Enhance classification save response handling and state persistence 12:18 Enforce PDF type domain isolation in classification to prevent structural pages matching architectural categories 07:31 Bypass AI vision calls on known pages to eliminate proxy 502/504 timeouts during classification 02:12 Fix upload state preservation so both structural and architectural files are retained 02:06 Truncate text arrays in upload.py to prevent MySQL payload overflow and HTTP 502 error 02:01 Split drawing uploads sequentially to eliminate HTTP payload timeouts and 502 Bad Gateway 01:56 Enable SQLite WAL mode and 60s timeout to prevent database locks and 502 errors 01:50 Eliminate RAM overhead during classification to fix HTTP 502 Bad Gateway OOM 01:49 Fix Dashboard create project button response handling 01:46 Optimize PDF page extraction and classification speed with 0% impact on extraction accuracy 01:43 Fix all remaining res.json calls in Workflow.jsx 01:40 Fix unexpected end of JSON input error in frontend response parsing Jul 25 01:43 Update qto_users schema and seed admin users in SQLite