Vision-Aid presented innovative approaches to digital accessibility and responsible AI at the 9th Annual Assistive Technology Conference at IIT Delhi, held from 2–4 October 2026. From helping blind users navigate inaccessible digital interfaces to embedding human judgement into AI-driven software development, the presentations highlighted how technology can advance digital inclusion, independence and opportunity for people with visual impairments.
Advancing Accessibility Through Artificial Intelligence
Artificial intelligence is creating new possibilities for people with visual impairments, from improving access to digital interfaces to opening pathways into technology-driven careers. At EMPOWER 2026, Vision-Aid highlighted the importance of developing AI solutions that address real-world accessibility barriers while keeping people, experience and human judgement at the centre of innovation.
The conference featured two complementary presentations by Vision-Aid team members. Together, they explored how AI can make digital environments more accessible and why human oversight remains essential to ensuring that technology works meaningfully for everyone.
1. AI Vision Assistant: Bringing Inaccessible Digital Interfaces Within Reach
Presented by: Gaurav Kumar
Co-authors: Manikandan Dhanakodi and Arunkumar Kumaragurubaran
Traditional screen readers such as NVDA rely on accessibility information provided by applications. When buttons, icons or other controls lack appropriate labels, or when interfaces are built using custom graphics or accessed through remote desktops, these tools may not identify the controls users need to interact with.
Vision-Aid’s AI Vision Assistant addresses this challenge by using artificial intelligence to interpret what is visibly displayed on a computer screen.
The Windows application captures the visible desktop and uses a vision-language model to identify interactive elements such as buttons, text fields, checkboxes, links and icons. It assigns each element a label, type and screen position, then presents the information through a keyboard-navigable overlay that works alongside NVDA.
Users can navigate the identified elements using their keyboard and press Enter to activate the corresponding on-screen control. The application also includes saved scans, quick scanning options and a dedicated shortcut for describing image-based CAPTCHAs.
An important technical contribution was the correction of screen coordinates across different resolutions and DPI scaling settings, helping ensure that keyboard-selected elements correspond to the correct controls on screen.
Testing, feedback and practical accessibility
The team tested the application in two stages: an initial closed group of three visually impaired testers, followed by a beta round involving ten visually impaired users. Participants reported positive experiences, including being able to interact with elements on websites that had previously been difficult or impossible to use with a screen reader alone.
The team also acknowledged areas requiring further development. Detection reliability varies across applications, response times can be challenging on busy screens, and formal quantitative measurements of accuracy and latency remain future priorities.
AI Vision Assistant Version 1.0 for Windows is publicly available at aivisionassistant.visionaid.org, where users can also provide feedback.
By supplementing existing assistive technology rather than replacing it, the application aims to reduce barriers in everyday digital tasks, workplace software and online interactions, supporting greater independence for people with visual impairments.
2. Beyond Automation: AI Accelerates, Human Guides
Presenters: Megha Patangi, Disha Jain and Parth Kohli
While AI can accelerate software development, generate code and automate accessibility checks, automation alone cannot guarantee that a digital product is genuinely accessible.
The presentation Beyond Automation: AI Accelerates, Human Guides explored the relationship between artificial intelligence and human responsibility across the Software Development Life Cycle (SDLC). Its central message was clear: AI can strengthen accessibility work, but human judgement remains essential to ensuring that digital experiences meet the real needs of people with disabilities.
A human-in-the-loop approach to accessibility
The paper proposed a conceptual framework that incorporates human-in-the-loop (HITL) checkpoints across five stages of software development:
- Planning: Identifying accessibility requirements and incorporating inclusion into user stories from the beginning.
- Design: Considering different user needs, assistive technologies and interaction patterns.
- Development: Reviewing AI-generated code and interfaces for meaningful accessibility.
- Evaluation: Combining automated testing with human assessment and assistive technology experience.
- Release: Maintaining accountability for accessibility outcomes before a product reaches users.
The framework focused on four essential dimensions of human intervention: context, intent, experience and accountability.
Practical examples included AI-generated student registration forms and automated image accessibility evaluations using axe Intelligent Guided Tests. These examples demonstrated why AI-generated outputs and automated findings need contextual interpretation and human verification before they can be considered genuinely accessible.
Moving beyond technical compliance
Automated tools and large language models can identify many accessibility issues, but they cannot independently establish whether every interaction is understandable, usable and appropriate for people with diverse disabilities.
For screen reader users, an interface may appear technically compliant while still presenting confusing navigation, unclear instructions during an actual task.
The paper advocated shifting accessibility earlier in the development process, rather than treating it as a final-stage correction. This approach can help organisations build inclusion into the foundations of digital products and reduce the risk of inaccessible experiences reaching users.
The presentation was appreciated by accessibility practitioners and conference attendees, prompting discussions about maintaining human accountability as software development becomes increasingly automated and agent-driven.
3. Extending the Vision: Building Career Pathways for Visually Impaired Talent in the AI Economy
Beyond accessible products, an equally important question is how people with visual impairments can participate in and benefit from the growing AI economy.
Vision-Aid’s proposed Ability Fellowship reflects this broader ambition. Designed as a one-year programme, the proposal aims to prepare visually impaired talent for professional opportunities through structured learning, mentorship and practical project experience.
The proposed programme includes three learning tracks:
- Builder: Developing skills in AI-powered applications, software development and agentic automation.
- Accessibility: Preparing participants for digital accessibility testing, auditing and consulting.
- Digital Marketing: Building capabilities in SEO, social media, performance marketing, analytics and AI-enabled workflows.
The proposed structure combines six months of instructor-led learning with six months of mentored project work. It also emphasises communication, professional development, assistive technology proficiency and the ability to verify AI-generated outputs.
A Common Vision: Technology That Works for People
Although the presentations addressed different challenges, they shared a common principle: meaningful accessibility requires more than technological capability. It requires understanding people, removing barriers and creating opportunities for independent participation.
AI Vision Assistant demonstrates how AI can help bridge gaps left by conventional assistive technologies. Beyond Automation: AI Accelerates, Human Guides highlights why human judgement must remain central to accessible software development. The proposed Ability Fellowship extends this thinking to education, skills development and professional opportunity.
Together, these efforts reflect Vision-Aid’s commitment to advancing accessibility through technology, practical innovation and the development of skills that support greater independence and participation.
As AI becomes increasingly integrated into everyday life, ensuring that its benefits are accessible to people with visual impairments is both a technical responsibility and an opportunity to build a more inclusive digital future.
