NatBraille is an open-source web editor for scientific documents in print and Braille, developed at the LIRIS laboratory (University of Lyon 1 / CNRS). It serves as the foundation upon which COOBRA builds to expand collaborative features.

• Official website: https://natbraille.univ-lyon1.fr/

Current features :

• Input and editing of scientific documents (text, equations, simple tables)
• Real-time transcription from print to Braille

Ongoing developments in COOBRA:

• Synchronous collaborative editing (multiple users, sighted and visually impaired)
• Multimodal awareness mechanisms (audio cues, haptic feedback)
• Advanced support for formulas 

ARCKit (ARC Knowledge & Inspection Toolkit) is a platform for the diagnostic evaluation and discovery of multimodal STEM accessibility systems. Developed as part of the COOBRA project, it puts into practice the findings of systematic reviews and empirical studies conducted by the team.

The platform is designed to be accessible to both non-specialists and researchers.

• To find the right tools: a simple rating system distinguishes between integrative tools (seamless use, low cognitive load) and additive tools (basic functionality but potentially tiring to use). Summaries in plain French accompany each entry.
• For evaluation and comparison: a diagnostic grid allows designers to position their system relative to a benchmark of 66 reference tools (2025 study). Results are visualized as radar charts.
• Exploring the corpus: a searchable database with filters (domain, platform, study type) provides access to all evaluated systems.
•  

Access and References

• Online platform: [link coming soon]
• Source repository: [link coming soon]
• Citation : ARCKit is a tool developed as part of the ANR COOBRA project, documented in Sadallah & Encelle (CHI ’26, W4A ’26).

CoobraTAB (Conversational Object-Oriented BLV Reasoning Agent for TABles) is an interactive system that enables blind or visually impaired people to explore and analyze complex data tables (financial, epidemiological, scientific) through natural language dialogue.

The system accepts tables in various formats (HTML, scanned images, PDF, CSV) and reconstructs their complete semantic structure (hierarchical headers, merged cells, aggregation relationships). Users can ask questions such as “Which are the three companies with the highest revenue?” or “Compare the results for France and Germany across all quarters.”

Main Features:

• Multi-format ingestion (HTML, images, PDF, CSV) with automatic recognition of table structures.
• Multi-turn dialogue with understanding of references (“and what about Q3?”) and analytical intent (comparison, filtering, trend calculation).
• Dynamic adaptation to the user’s profile: level of expertise, type of visual impairment, exploration style, desired level of detail.
• Voice output optimized for screen readers (JAWS, NVDA, VoiceOver, TalkBack) and mathematical Braille formatting.
• Generation of responses grounded in the actual data of the table (no hallucinations) thanks to a grounding mechanism.

Status: Prototypes currently under development. A user evaluation protocol is planned in several phases (formative, technical, cross-format comparative).

Access: (link coming soon – open-source repository and online demo)