Alpha Forge — an agentic AI operating system for systematic trading.

Feature engineering, labeling, alternative bars, and leakage-safe datasets for financial ML.

ML-based buy signal detector for Tehran Stock Exchange using XGBoost & Random Forest

38M-param time-series world model: FSQ tokenizer → Mamba-2 JEPA → OT-CFM → TD-MPC2 agent. 838M tokens, TPU v6e, JAX/Flax.

End-to-end ML pipeline that predicts BTC/USDT price direction (4h horizon) using XGBoost + Optuna + SHAP. 9-phase architecture, Walk-Forward Validation across 15 folds, 37 technical indicators, 98 automated tests. ROC-AUC: 0.5431.

Predict S&P 500 stock performance using a graph neural network that models market correlations and sector relationships to generate long-short portfolio signals.

Heterophily-aware GNN pipeline for anti-money laundering detection. H2GCN + XGBoost cascade achieves PR-AUC 0.595 on IBM AML dataset, 119× improvement over standard GAT.

Graph + temporal deep learning for cross-sectional S&P 500 ranking. 9-variant ablation, 224 tests, val IC 0.0284 on yfinance data.

Deep RL agent for financial market signal generation — PPO/A2C/SAC/TD3, 99 indicators, ensemble signals, 4-level quality gate

Deep learning pipeline for financial time-series forecasting using LSTM, CNN, CNN–LSTM and ResNet–LSTM with Gramian Angular Difference Field (GADF) encoding and an interactive Streamlit dashboard.

NIFTY 50 5-day trend classification using Decision Tree, Random Forest and Logistic Regression with live prediction system.

AI-powered loan approval prediction system using XGBoost with 96.3% accuracy. Predicts credit eligibility based on income, credit score, DTI ratio & 20+ financial features. Built with Python, Scikit-learn & Streamlit

FinFusion: S&P 500 return forecasting with Temporal Fusion Transformers - compares TFT, ARIMAX, LSTM, and regime-aware variants.

QuantLab alpha construction component for purified thematic signals, walk-forward weighting, IC evaluation, turnover diagnostics, and ML alpha experiments.

Advanced ML system combining LSTM attention networks, Transformer architectures, and gradient boosting ensembles for financial time series forecasting

Reinforcement-learning aggregator on top of the TradingAgents multi-agent LLM trading framework. PPO policy that beats buy-and-hold on the 2026 YTD test across two LLM backbones (Anthropic + OpenAI); cross-LLM transfer holds. Course project, Columbia IEORE 4733.

Real-time financial sentiment pipeline: 3-model FinBERT ensemble (86.7% acc, neg PR-AUC=0.908) · MC Dropout uncertainty · TimescaleDB · Kafka · FastAPI · MLflow · Prefect 2

Transformer‑based Bull/Bear classifier for Bitcoin using long‑window trend features and pretrained inference‑only weights.

...

In-progress AI-assisted systematic alpha research platform for factors, signals, portfolio construction, backtesting, and research automation.