File size: 2,511 Bytes
00bfe7c
cf91278
00bfe7c
 
cf91278
 
 
 
 
 
33ef360
cf91278
 
 
 
 
 
 
03bff2e
cf91278
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
---
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>