Skip to content

PRD.md

Latest commit

 

History

History
222 lines (159 loc) · 11.6 KB

PRD.md

File metadata and controls

222 lines (159 loc) · 11.6 KB

Product Requirements Document (PRD)

Accellens — AI Accessibility Compliance Platform

Version: 2.0 Update Date: November 23, 2025 Status: MVP in development (Stage 1 completed, transitioning to Stage 2) Product: Accellens (AI Accessibility Compliance Platform) Developer: Tridention

1. Introduction

1.1 Description

Accellens is a cloud-based AI platform that performs deep audits of web and mobile interfaces for compliance with accessibility standards (WCAG 2.2, EN 301 549, ADA Section 508). The solution combines:

  • Automatic analysis of code, DOM structure, and visual layer;
  • Modeling of real user scenarios;
  • Emulation of screen reader behavior and keyboard navigation;
  • AI interpretation of visual and cognitive perception of interfaces.

1.2 Goal

To make the accessibility of digital products measurable, visual, and predictable, reducing the gap between technical audit and the real user experience of people with disabilities.

2. Core Product Objectives

  1. Perform automatic audits of web and mobile interfaces for compliance with WCAG 2.2 / EN 301 549 / ADA Section 508.
  2. Emulate interactions of users with various disabilities:
    • Blind user (NVDA, JAWS, VoiceOver, TalkBack).
    • Limited mobility (keyboard only / on-screen keyboard).
    • Low vision (contrast, large font, magnification).
    • Cognitive difficulties (content structure, predictability, load).
  3. Generate reports with recommendations: severity, impacted users, auto-fixes, and justified priorities.
  4. Integrate into existing processes (CI/CD, IDE, browsers, bug trackers).
  5. Train AI models on real interaction scenarios and synthetic data.

3. Target Audience

User Type Role Goal
QA Engineers Accessibility check before release Accessibility test automation, compliance assurance
Developers Fixing identified issues Obtaining technical descriptions of problems with code fix examples
Accessibility Specialists Deep analysis and reporting Monitoring standards and preparing audits
Product Managers UX evaluation for all user categories Understanding risks, priorities, and impact on business
Gov and Corporate Clients Compliance with regulations Avoiding legal risks and fines
Enterprise Clients On-premise deployment Installing the solution on their own servers for security, compliance (GDPR, HIPAA, FedRAMP) or data residency

4. Competitors and Uniqueness

Competitors: Deque Axe, Siteimprove, Pa11y, Level Access, EqualWeb, Evinced.

Accellens Unique Advantages:

  • Real perception condition modeling: AI user simulations instead of pure static HTML analysis.
  • Mobile application support (iOS/Android) alongside web interfaces.
  • Accessibility Accuracy Score (AAS) — a comparative accessibility metric updated by AI.
  • Support for real devices and screen readers via cloud-runner infrastructure.
  • Constant model training on real interactions of people with disabilities (anonymized data + synthetics).
  • On-premise deployment (v2+): the ability to install the solution on client servers for enterprise clients with security, compliance (GDPR, HIPAA, FedRAMP), or data residency requirements.

5. Product Architecture (High Level)

Frontend

  • Web Dashboard on Next.js: Project Dashboard, Test Results, Reports, Recommendations, Integrations, Settings.
    • Desktop-first approach: the interface is designed and optimized primarily for desktop devices (≥1024px), then adapted for tablets and mobile devices while maintaining full functionality.
    • Responsive design: all functions are available on all devices (desktop, tablet, mobile) with adapted UI for each screen size.
    • Touch-friendly: on mobile devices, all interactive elements have a minimum size of 44x44px for convenient use.
    • Detailed responsive design requirements are described in docs/development/design-system.md.
  • Mobile Client (React Native / Flutter) for viewing reports and quick audits (roadmap).

Backend

  • Node.js (API Gateway): REST and GraphQL API, authentication, authorization, multi-tenant data isolation.
  • Python Microservices (audit, AI pipelines, simulations).
  • PostgreSQL (metadata, results): multi-tenant data structure with organization-level isolation.
  • Redis / RabbitMQ (task queues).
  • S3-compatible Object Storage (artifacts, screenshots, audio): isolated storage for each organization.

Multi-tenancy:

  • Each company has its own Organization, which contains multiple Projects.
  • All data is isolated at the organization level (organization_id in every table).
  • Users see only their organization's data.
  • RBAC at two levels: Organization level (owner, admin, member) and Project level (owner, maintainer, auditor, viewer, integrator).

AI Layer

  • LLM for error interpretation, generating recommendations and explanations.
  • Vision model for visual contrast, textures, and UI component analysis.
  • Audio module for simulating speech interfaces (NVDA, VoiceOver).
  • Reinforcement Learning for optimizing test scenarios and improving technical problem description quality.

Cloud Execution

  • Dockerized environment with headless browsers (Chromium, Safari, Firefox).
  • Android and iOS emulators (XCUITest, Espresso, Appium).
  • Running accessibility tests via API / CLI / CI pipeline.

6. Main Modules

Module Description
Audit Engine Analysis of code, DOM, styles, events; WCAG 2.2, EN 301 549 rules checking
AI User Simulation Behavior of users with screen readers, keyboards, vision impairment
Cognitive Load Analyzer AI assessment of content structure and readability, cognitive load
Contrast & Visual Layer Analyzer Colors, contrast, alternative text, scaling checks
Reports & Fix Suggestions Reports, severity, impacted users, Accessibility Accuracy Score, technical problem descriptions with code fix examples
CI/CD Integrations Support for Jenkins, GitHub Actions, GitLab CI, Azure DevOps, Cypress
API & SDK REST/GraphQL API, SDK (JS/TS, Java, Python), and IDE/browser plugins
Learning Engine Self-learning of models on audit data and user interactions

7. User Scenario (Example)

  1. QA or developer connects a site (URL) or uploads a mobile app build.
  2. Accellens launches multi-context audit (WCAG, EN, ADA).
  3. AI models the use of VoiceOver and keyboard navigation, records cognitive load.
  4. An interactive report is generated:
    • Severity matrix (Critical / Major / Minor).
    • Affected user segmentation.
    • Technical problem description and code fix example.
    • Accessibility Accuracy Score.
  5. Results are published to CI/CD, sent to Jira / bug tracking, and available in the dashboard.

8. Success Metrics (KPIs)

Metric Target
Average audit time (10 pages) < 3 minutes
AI model accuracy (vs manual audit) ≥ 90%
Share of auto-generated recommendations ≥ 80% of cases
Time to implement fixes -30% from baseline
User UX rating > 4.5 / 5

Additionally: Pilot client retention ≥ 80%, NPS ≥ 50.

9. Roadmap

MVP (First 6 Months) — Stage 1 completed (November 23, 2025)

Implemented:

  • ✅ Web page audit (WCAG 2.2 AA) — Audit Engine applies rules from axe-core.
  • ✅ Screen reader emulation (text layer) — basic text simulation implemented.
  • ✅ Reports with recommendations and machine explanations — PDF/JSON/SARIF reports are generated, AI explanations for ≥80% findings.
  • ✅ Web dashboard + REST/GraphQL API — basic dashboard and full API implemented.
  • ✅ GitHub Actions integration — GitHub Action integration works (dry-run).

In Progress (Stage 2):

  • ⬜ Increase test coverage to target values (backend ≥80%, frontend ≥75%).
  • ⬜ Advanced dashboard features and analytics.

v1.0 (6–12 Months)

  • Mobile application support (iOS/Android).
  • Visual and contrast analysis (Vision AI).
  • Cognitive AI assessment (structure, cognitive load).
  • EN 301 549, ADA support, and SARIF export.
  • CI/CD integrations (GitLab, Jenkins) and Jira.

v2.0 (12–18 Months)

  • Realistic VoiceOver/TalkBack simulation (audio + gestures).
  • Live Replay of user sessions (anonymized).
  • AAS (Accessibility Accuracy Score) model with dynamic weights.
  • Fix Suggestions Engine — technical problem descriptions with code fix examples.
  • Audit template marketplace (extensibility).
  • Trainable model on client data (privacy-preserving).

Detailed backlog and exit criteria by stage — see roadmap.md.

10. Risks and Limitations

Risk Possible Solution
Screen reader emulation limitations Integration with open-source bridge, real-device cloud, voice layer
High load during UI rendering Container pool with auto-scaling, task queue
ADA/EN licensing restrictions Consultations with digital accessibility lawyers
AI interpretation accuracy Human-in-the-loop QA, regular retraining
Data privacy ISO27001 / GDPR-compliant storage, redaction pipeline

11. Business Model

  • Freemium: up to 3 pages/scans free (limited checks).
  • Pro (SaaS): $49–$299/mo depending on volume, integrations, and SLA.
  • Enterprise (SaaS): custom SLA, white-label, extended support.
  • Enterprise (On-Premise): custom SLA, white-label, self-hosted/on-prem deployment, licensing per organization/user/scan, air-gapped mode support.
  • API Licensing: license for large companies and CI integrators.

12. Related Documentation