Bidh: AI-Powered Scots Gaelic Language Tutor

The Brief

Scottish Gaelic is a minority language with limited access to native speakers for practice. Learners with strong reading ability often lack opportunities for spoken conversation, and existing language tools rarely support Gaelic at all - let alone with the nuance of dialect, grammar correction, and natural speech.

The Challenge:

• No existing AI-powered tools for spoken Gaelic practice
• Learners need both immersive conversation and structured correction
• Gaelic's orthography, grammar, and regional dialects require specialist handling
• The solution needed to feel like a conversation, not a language drill

The Strategy

Build a real-time voice interaction system that acts as both a conversation partner and a language tutor. The system needed to understand spoken Gaelic, respond naturally in Gaelic, and switch seamlessly between immersive conversation and structured teaching when the learner makes mistakes.

The Approach:
Edge computing keeps latency low enough for natural conversation. By distributing processing across local hardware and cloud services, the system achieves response times that feel conversational rather than transactional - critical for language learning where hesitation breaks the flow.

The Execution

Voice Pipeline
Built a real-time audio pipeline handling speech capture, transcription, language model reasoning, and voice synthesis. The system processes spoken Gaelic and responds with a natural Gaelic voice, maintaining conversation context across the full session.

AI Integration
Integrated large language model capabilities with carefully engineered prompting to handle Gaelic's specific grammatical structures. The system operates bilingually - conversing in Gaelic while providing corrections and explanations in English when teaching.

Dual Mode Interaction
Designed two switchable modes: Chat mode for immersive Gaelic conversation, and Tutor mode where the system detects errors, explains the grammar in English, demonstrates the correct form in Gaelic, and drills the learner with fresh examples using the same structure.

Edge-First Infrastructure
Deployed on Cloudflare Workers with Durable Objects for session persistence, keeping the orchestration layer at the edge. This hybrid approach balances performance-critical processing with cost-effective cloud services.

Gaelic-Specific Engineering
Addressed the unique technical requirements of Scottish Gaelic: UTF-8 handling for grave accents throughout the stack, dialect-aware responses, and latency management tuned for the natural pauses in Gaelic speech patterns.

The Result

Bidh provides a functional, voice-driven Gaelic tutor that can hold a natural conversation and teach at the same time. Crucially, the system is designed to get better with use - corrections from native speakers feed back into the platform, and learner interactions inform where the teaching approach needs refinement.

Outcomes:

• Real-time voice interaction with response times suitable for natural conversation
• Bilingual teaching: Gaelic immersion with English explanations when needed
• Structured error correction that reinforces grammar through practice
• Edge-deployed architecture keeping costs low and latency manageable
• A platform designed for expansion into dialect support, progress tracking, and curriculum-based learning
• A feedback loop planned for native Gaelic speakers to evaluate and correct the system's output, ensuring language quality improves over time rather than relying solely on AI-generated content