Current Focus:
- Building AI-driven financial analysis tools leveraging LLMs and reinforcement learning
- Training small-scale language models for specialized reasoning tasks
- Exploring Group Relative Policy Optimization (GRPO) for model alignment
- Developing mathematical visualization tools with Manim
Expertise Areas:
- CFA training, Derivatives, and Options Trading
- Congressional Trading Analysis and Market Microstructure
- Quantitative Finance and Risk Management
- Machine Learning Engineering (PyTorch, TensorFlow)
- Mathematical Visualization and 3D Graphics
Unique Expertise
- Advanced PyTorch implementations
- Reinforcement Learning from Human Feedback (RLHF)
- Large Language Model fine-tuning and alignment
- Computational geometry and fractal mathematics
I once got Karpathy to reply "Nice."
My meager Google Scholar reference
3D visualization of options volatility skew across strikes and maturities. This surface represents the Black-Scholes implied volatility smile, crucial for pricing exotic derivatives and understanding market risk premia. |
A fractal Archimedean solid demonstrating computational complexity in 3D geometry. Each vertex recursively contains another complete polyhedron, requiring precise transformations across 62 unique faces - a challenging test case for LLM mathematical reasoning. |
- Volatility Surface: Real-time Black-Scholes implied volatility calculations
- 3D Rendering: Custom Manim animations for financial mathematics
- Computational Geometry: Precise coordinate transformations in 3D space
- Data Visualization: Market data analysis and pattern recognition
As of Nov 2025: 9 models / 6 datasets / 6 Spaces. Featured models and datasets:
| Model | Focus | Base | Size | Updated |
|---|---|---|---|---|
| Qwen3-0.6B-Dakota-Grammar-RL | Dakota language grammar via GRPO | Qwen3-0.6B | 0.8B | 1 day ago |
| nanochat-AquaRat | RL training on algebraic reasoning (AQuA-RAT) | nanochat | - | 18 days ago |
| nanochat561 | Text generation experiments | nanochat | - | 20 days ago |
| Qwen.5B-OpenR1Math | Math reasoning (Open-R1 style) | Qwen2.5-0.5B-Instruct | 0.5B | Feb 13 |
| Qwen.5B-GSM8K | Math finetune (GSM8K emphasis) | Qwen2.5-0.5B-Instruct | 0.5B | Feb 12 |
| GRPOtuned / GRPOtuned2 | GRPO experiments on 0.5B LLMs | Qwen2.5-0.5B-Instruct | 0.5B | Feb 6-9 |
| Dataset | Description | Size | Updated |
|---|---|---|---|
| dakota-bilingual-qa | Dakota-English bilingual Q&A pairs | 2.45k rows | 5 days ago |
| Stoney10kRL | Stoney Nakoda RL training data | - | 13 days ago |
| synthetic_stoney_data | Synthetic Stoney Nakoda language data | 68.8k rows | Apr 28 |
| StoneyNakoda45k | Stoney Nakoda language corpus | - | Apr 5 |
| StoneyNakoda | Stoney Nakoda language dataset | 14.5k rows | Jan 22 |
| StoneyCIL | Stoney Nakoda CIL dataset | - | Jan 4 |
Direct links:
- Models: Qwen3-0.6B-Dakota-Grammar-RL, nanochat-AquaRat, nanochat561, Qwen.5B-OpenR1Math, Qwen.5B-GSM8K, GRPOtuned, GRPOtuned2
- Datasets: dakota-bilingual-qa, Stoney10kRL, synthetic_stoney_data, StoneyNakoda45k, StoneyNakoda, StoneyCIL
- Spaces: Dakota Grammar RL Demo, deepsitecoder, Stoney-1, Ask About Stoney
Quickstart (try a model in 5 lines):
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
mdl = "HarleyCooper/Qwen3-0.6B-Dakota-Grammar-RL" # or any model ID above
tok = AutoTokenizer.from_pretrained(mdl)
model = AutoModelForCausalLM.from_pretrained(mdl, torch_dtype=torch.float16, device_map="auto")
print(tok.decode(model.generate(**tok("Translate to Dakota:", return_tensors="pt").to(model.device), max_new_tokens=64)[0], skip_special_tokens=True))
