Égua Assist
Accessible platform that teaches programming logic to visually impaired students — voice interaction, NLP, and LLM-generated explanations.
Overview
Égua Assist is an accessible web platform that teaches programming logic to visually impaired students. The IDE is driven by voice: users speak commands and questions, and the system replies with spoken explanations and generated code, powered by a set of LLM microservices with a speech layer.
index.html.
Features
- Voice-first editor — screen-reader-friendly interface where users command the IDE by speech.
- LLM services — independent Python microservices for code generation and explanation (Gemini, GPT-4o-mini, Llama-3, Phi-2, Qwen2), plus Whisper for speech.
- Modular design — each capability is its own service with its own port, so models can be swapped.
- Static frontend (HTML/JS) with no build step.
Tech stack
- Python (Flask/FastAPI) microservices, torch/torchtext for local models
- PHP (phpdotenv) for serving the static frontend
- LLM providers: Gemini, OpenAI, local models via Ollama-style APIs
Running locally
# each service runs on its own port, e.g.:
python servers/code_generation/gemini_code_generator.py
php -S localhost:8080 # serves the IDE
Requires API keys in .env for the LLM services (OPENAI_API_KEY, GOOGLE_API_KEY).
Why this project?
Accessibility is a genuine engineering challenge — the constraints (no visual feedback, voice-first interaction) force careful design of every interaction, and the multi-model architecture shows how to integrate several LLM backends behind a single interface.