Documentation
EchoGuide — Bilingual Voice Assistant for Accessible Android Interaction
EchoGuide is an accessibility-focused Android voice assistant designed to help blind and low-vision users interact with mobile applications through natural spoken commands. Instead of requiring users to manually locate and operate individual interface elements, EchoGuide allows them to describe what they want to accomplish in Amharic or English, then intelligently plans, validates, and executes the required actions on their behalf.
The system combines speech recognition, AI-powered task planning, Android Accessibility Services, and text-to-speech into a single voice-driven interaction pipeline. A user's request is captured on the device, converted into text, and processed together with contextual information about the current screen. The AI then generates a structured action plan consisting of safe operations such as tapping, scrolling, entering text, navigating back, or returning to the home screen. Before execution, the plan is validated both by the backend and on the device to ensure that only permitted actions are performed.
A key aspect of EchoGuide is its Amharic-first approach to accessibility. The platform supports both Amharic (am-ET) and English (en-US) across speech recognition, AI planning, and voice responses, making AI-powered mobile accessibility more relevant to Ethiopian users and other Amharic-speaking communities.
The application is built using a React Native and Kotlin architecture. React Native provides the application interface and user experience, while the critical voice-processing and accessibility pipeline is implemented natively in Kotlin. This separation enables EchoGuide to interact reliably with other Android applications while keeping the latency-sensitive voice pipeline independent of the JavaScript bridge.
Safety and privacy are core design principles. EchoGuide restricts AI-generated plans to a small set of predefined actions, validates commands before execution, requires explicit spoken confirmation for destructive operations, and provides spoken feedback for every system state. Audio and transcripts are not stored by default, and users can control which applications EchoGuide is permitted to interact with.
The backend follows a modular architecture built with TypeScript, Express, PostgreSQL, Redis, and OpenAPI, while Addis AI provides the speech-to-text, language-processing, and text-to-speech capabilities. The project also incorporates automated unit and integration testing, speech-recognition evaluation, Kotlin pipeline testing, API contract validation, and documented architecture decisions.
Technology Stack: React Native, Kotlin, TypeScript, Node.js, Express, PostgreSQL, Redis, OpenAPI, Expo, Android Accessibility Services, and Addis AI.
Core Focus: AI agents, mobile accessibility, voice interfaces, natural-language interaction, Amharic language technology, Android automation, privacy, and safe AI execution.
