Skip to content

MergeMark Beta (v0.9.2)

Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 30 Jul 20:39
v0.9.2
b4dc376

Welcome to MergeMark v0.9.2! This update brings significant improvements to our ingestion engine, ensuring crystal-clear diagram rendering, as well as highly requested features like Anki export and custom taxonomy management.

✨ What's New

  • Anki Import & Export: You can now seamlessly export your flashcards directly to Anki, making spaced repetition studying more connected than ever.
  • Custom Subject Taxonomy: Take control of your repository organization. The new Taxonomy Manager allows you to define custom subjects and topics, tailoring the workspace exactly to your curriculum needs.
  • Interactive Diagram Zoom (Lightbox): Diagrams and figures now support a click-to-enlarge lightbox. Even if a question card shrinks an image to fit your screen, you can open it to its native high-resolution size to inspect tiny details like axis labels or circuit values.
  • High-DPI Document Rendering: We've overhauled the document rendering pipeline. Pages are now rendered at ~200 DPI (over 3x the previous resolution) with improved JPEG quality. Say goodbye to blurry subscripts and mushy line-art.
  • Rich Text & UI Polish: We've ironed out several UI quirks in the Rich Text Editor and repository modals for a smoother editing experience.

🔧 Technical Improvements

  • Vector-Aware PDF Processing: The ingestion pipeline now intelligently detects vector paths (lines, curves, strokes) in addition to raster images. Physics and mechanics pages containing native PDF drawings will no longer be bypassed by the text-only fast path.
  • Vision API Optimization: Image payloads now request detail: high for OpenAI-compatible vision models, ensuring the AI can read complex diagrams without downsampling.
  • Expanded Context Windows: We've doubled the max_output_tokens budget for PDF vision parsing (from 16k to 32k) to prevent output truncation on lengthy question derivations.
  • Foundational Pipeline Prep: The rendering engine overhaul lays the groundwork for Phase 1 (sub-page splitting and MCQ slicing) by guaranteeing accurate pixel-per-page coordinates.

🐛 Bug Fixes & Refactors

  • Sentinel Image Corruption: Fixed a critical bug where the literal string 'TEXT_ONLY' was being sent as a base64 image during multi-page chunks, which corrupted API requests and caused model refusals.
  • UI & Styling: Updated the app logo background color and resolved minor alignment bugs across the repository feed.
  • Documentation: Updated the project README with comprehensive features, tech stack details, and an installation guide.

Full Changelog: v0.9.0...v0.9.2

Assets 9
Loading