Project: Speaking
Project: Speaking is an audio-first, conversational AI language practice platform engineered to turn passive knowledge into instinctive speaking reflexes. Traditional language learning often traps learners in endless passive listening and multiple-choice drills. Project: Speaking places active verbal production at the core of every session, challenging you to produce high-stakes, instantaneous oral responses with realistic pause-and-anticipation pacing. Step into over 27 curated situational roleplays or craft completely custom scenarios—from high-stakes tech job interviews and salary negotiations to everyday travel emergencies and casual café banters. With integrated Spaced Repetition (SRS), every key phrase and vocabulary item you capture during live lessons is automatically scheduled for smart oral review sessions, ensuring permanent long-term retention. Engineered with a Hybrid Dual AI Engine, Project: Speaking provides seamless one-click switching between ultra-low latency Google Gemini Live cloud streaming and a 100% private, on-device local C++ pipeline (Metal GPU accelerated on Apple Silicon) powered by Qwen 3.5 LLM, OpenAI Whisper Large-v3 Turbo STT, and Kokoro-82M neural TTS. All user data, custom scenarios, and sentence vaults remain 100% local in an isolated SQLite database.
Engineering Highlights & Capabilities
Active Verbal Production
Designed around active spoken response loops rather than passive multiple choice. Builds genuine oral fluency and reflex memory.
Hybrid Dual AI Engine
Instant one-click switching between Google Gemini Live low-latency cloud audio and 100% offline Metal GPU C++ sidecars.
Sentence Vault & SRS
Capture challenging phrases directly during conversation. Spaced Repetition algorithms schedule smart oral review flashcard drills.
100% Local-First Privacy
All learning history, voice profiles, and custom scenarios stay in an isolated local SQLite database. Zero cloud vendor lock-in.