My Recent Projects I have Built:
Stack Society HUD
A PokerNow analytics platform for exploring sessions, replaying hands, and understanding player tendencies.
- Next.js
- React
- TypeScript
- Node.js
- Python
- PostgreSQL
- Prisma
- Tailwind CSS
- Docker
DISTILBERT / SST-2
Prompt Compression
Keep the query. Compress the context.
| Method | Mean tokens | Accuracy | ms / example |
|---|---|---|---|
| Full prompt | 136.16 | 90.44 ± 0.48% | 0.351 |
| IDF · 25% demo budget | 97.16 | 90.90 ± 0.43% | 0.179 |
43 configurations · 129 fits · 3 seeds
Accuracy: mean ± seed SD. Warm GPU pipeline timings, batch 16.
Small accuracy differences do not establish consistent superiority.
Prompt Compression
An NLP study exploring how shorter prompts affect text classification accuracy, input length, and inference cost.
- Python
- PyTorch
- Transformers
- DistilBERT
- scikit-learn
- pandas
- NumPy
- IDF Compression
- Learned Compression
STRATEGY / OPPONENT MODELING
PokerNow Deep Learning Assistant
From table state to a more informed decision
The player makes the final decision
PokerNow Deep Learning Assistant
A poker decision-support tool combining CFR+/DCFR strategies, legal-action checks, and opponent-model diagnostics to help players evaluate their next move.
- Python
- TypeScript
- Node.js
- PostgreSQL
- Playwright
- scikit-learn
- pandas
- NumPy
- LightGBM
- XGBoost
- CatBoost
- Extra Trees
- CFR+
- DCFR