FinnewsHunter: Multi-agent financial intelligence platform powered by AgenticX. Real-time news analysis, sentiment fusion, and alpha factor mining.

Quantitative investment strategy capturing divergence between analyst ratings and news sentiment using FinBERT

ML-driven alpha signal generation for U.S. equities. Ridge, Random Forest, and XGBoost trained with walk-forward validation. XGBoost achieves Sharpe 1.07 out-of-sample. Full feature importance and IC stability analysis.

VLLM-based trading system

GPU-Accelerated NDI Broadcast Engine with Desktop Capture, NDI Main + Backup Input, Main Backup Failover and NDI + Decklink Out

ML-driven adaptive quantum error mitigation for NISQ devices. XGBoost selector chooses between Linear ZNE and Richardson ZNE per circuit/noise regime. 64.3% accuracy, +11.9pp over fixed-method baseline. Preprint: doi.org/10.5281/zenodo.20140750

Implemented and adapted selected alphas from the WorldQuant 101 Alphas framework to the cryptocurrency market. Built a 2020–2025 dataset with train/test split, evaluated signals using IC statistics, combined weak alphas via rank normalization, and backtested portfolio strategies, achieving improved Sharpe over an equal-weight benchmark.

Three-signal alternative data system (Reddit/FinBERT sentiment, Finnhub news, Sentinel-2 satellite imagery) with IC/ICIR evaluation. Best result: PLTR IC(1d)=+0.352, p<0.01.

v3min8e

Developed a comprehensive alpha generation and portfolio optimization system combining traditional factor models with alternative data sources, achieving consistent risk-adjusted outperformance over market benchmarks through sophisticated signal blending and dynamic risk management.

An Intraday Alpha Backtesting Framework for popular equities

Momentum trading strategy for crypto using ML (XGBoost) vs traditional logic — with backtesting, VaR risk analysis & Monte Carlo simulation.

Zero-infra alerter for Congress trades + SEC 13D/G + Asian 5%-stake disclosures. Runs on GitHub Actions cron, DMs Telegram.

A repository containing 8 alpha signals with a signal combination engine incorporate a risk model, with 76+ automated tests.