Pioneer in transformer-based mobile writing (2022) - Legacy Project
🏛️ Groundbreaking Achievement: One of the first iOS apps with transformer models trained from scratch on carefully curated datasets. Full GPT-style architecture running locally on mobile devices!
DeepWriter represented a massive leap from our Markov chain-based Creative Writer. This was one of the first implementations of transformer architecture on iOS, featuring models trained entirely from scratch on meticulously curated datasets spanning literature, technical writing, poetry, and contemporary prose.
All transformer inference happened on-device using CoreML optimization. No internet required, no data ever left the device - a revolutionary approach when cloud-based AI was the norm.
The app featured custom-trained transformer models optimized for different writing styles, with automatic model updates delivered via Amazon S3. Each model was carefully trained on domain-specific corpora to ensure high-quality, contextually appropriate suggestions.
Full transformer models trained from scratch, not fine-tuned. Each model carefully architected for specific writing domains.
Meticulously curated datasets from quality sources - no web scraping. Every text manually vetted for quality and appropriateness.
PyTorch → ONNX → CoreML conversion pipeline with custom optimizations for Apple Neural Engine acceleration.
Complete on-device processing. No telemetry, no analytics, no data collection. True privacy in the age of surveillance capitalism.
Different transformer models for poetry, technical writing, creative prose, and business communication - all swappable in real-time.
Models specifically designed for mobile constraints - 50M parameters optimized to run smoothly on A12 Bionic and newer.
DeepWriter represents a breakthrough in mobile AI - the first iOS app to successfully implement full GPT model inference on-device, ensuring both powerful AI assistance and complete user privacy.
Experience the power of local AI-powered writing
Demo Video Coming Soon