| --- |
| organization: AMFORGE |
| --- |
| |
| <div align="center"> |
|
|
| # AMEFORGE |
|
|
| **Independent AI Research Studio** |
|
|
| [Website](https://ameforge.tech) · [GitHub](https://github.com/Volgat) · [Google Play](https://play.google.com/store/apps/dev?id=8746471888242300768) |
|
|
| </div> |
|
|
| --- |
|
|
| ## About |
|
|
| Ameforge is an independent AI research studio. We design novel architectures, publish peer-reviewed research, and deploy production-ready models with software that brings these innovations directly to users. |
|
|
| Our work spans deep learning architecture research, natural language processing, time-series forecasting, and mobile software development. |
|
|
| --- |
|
|
| ## Research |
|
|
| ### EARCP — Ensemble Adaptive Recurrent Cascade Predictor |
|
|
| Published on arXiv · Validated on Kaggle Hull Tactical Market Prediction (Rank 708 / 18,000+ teams · Top 24%) |
|
|
| A multi-expert ensemble architecture combining CNN, BiLSTM, and Transformer components for adaptive time-series prediction. Applicable to financial forecasting, sensor data, computer vision, and autonomous systems. |
|
|
| ```bash |
| pip install earcp |
| ``` |
|
|
| 📄 [arXiv Paper](https://arxiv.org/abs/2603.14651) · 🏆 [Kaggle Leaderboard](https://www.kaggle.com/competitions/hull-tactical-market-prediction/leaderboard) |
|
|
| --- |
|
|
| ### SparseMind — Sparse Neural Architecture Research |
|
|
| Experimental research into high-sparsity neural architectures achieving 87.5% weight sparsity while maintaining competitive perplexity scores. Designed for efficient inference on resource-constrained environments. |
|
|
| --- |
|
|
| ### NexusBPE — Custom Tokenizer Pipeline |
|
|
| A custom Byte-Pair Encoding tokenizer trained on a curated multilingual corpus (code, natural language, technical documents). Designed to complement Ameforge's internal language model research. |
|
|
| --- |
|
|
| ### MemoryBank (LPOLMemory) — Continual Learning Research |
|
|
| Research into long-term memory mechanisms for neural networks, addressing catastrophic forgetting in continual learning scenarios. |
|
|
| --- |
|
|
| ## Models |
|
|
| All published models are available under the **CKL (Custom Knowledge License)**. Commercial licensing available — contact [contact@ameforge.tech](mailto:contact@ameforge.tech). |
|
|
| | Model | Task | Status | |
| |---|---|---| |
| | gc_editor1 | Text Generation | Active | |
| | gc_editor | Text Generation | Active | |
| | gearcut_tok | Tokenizer | Active | |
| |
| --- |
| |
| ## Contact |
| |
| 📧 [contact@ameforge.tech](mailto:contact@ameforge.tech) |
| 🌐 [ameforge.tech](https://ameforge.tech) |
| |
| --- |
| |
| <div align="center"> |
| <sub>© 2025 Ameforge. All rights reserved.</sub> |
| </div> |
| |
