I’m upgrading the Voice AI inside my “Revise like a Teacher” app and need it to go beyond simple speech-to-text. The engine must pick up the learning context behind a student’s words, recognise references in previous-year exam questions or textbook passages, and then respond as naturally as a real teacher would. The core objective is therefore enhanced understanding of context; raw recognition accuracy and voice quality are already acceptable. You’ll be working with curriculum-heavy material, so domain-adapted language models, embeddings or custom transformers that can digest educational phrasing are welcome. Whisper, DeepSpeech, Kaldi, PyTorch- or TensorFlow-based pipelines are all fine as long as latency stays low for mid-range devices. To move forward I need a detailed project proposal rather than a generic CV. In it, outline: • Your plan for collecting / formatting training data drawn from textbook chapters and past exam papers • The fine-tuning methodology and evaluation metrics you’ll use to prove contextual comprehension gains • Expected training time, model sizes, deployment approach and any privacy safeguards Deliverables 1. Fine-tuned model or API ready for drop-in use inside the Flutter app 2. Demo that interprets a sample textbook section plus a set of past questions, then answers voice queries context-sensitively 3. README with clear setup and reproducible test instructions Once I can see a solid path to measurable context awareness, we’ll kick things off immediately.