No description available

Tools for training, analysis and execution of an optimized task scheduling RL agent on GPUs with Multi-Instance GPU (MIG).

Based on GEN/ETH market microstructure LOB data, trades, objectives & constraints – recommend suitable algos which maximize pnl/minimize inventory losses within a fixed time horizon. The Reinforcement Learning (RL) method is chosen to help achieve optimized trade execution relative to given constraints.

♟️ Optimized Chess RL Trainer using DQN vs Stockfish. Built with PyTorch and python-chess, it learns using per-move rewards from Stockfish evaluations. Implements Prioritized Experience Replay (PER) and parallel CPU/GPU execution for faster training. The agent dynamically adjusts Stockfish skill level based on performance.

ResNet50-based deep learning model for multiclass classification of harmful brain activity using raw EEG (Parquet, 200 Hz) and regional spectrogram power (LL, RL, LP, RP). Trained with Stratified Group K-Fold for patient-wise generalization. Uses zero-imputation for stable tensor input. Optimized for Kaggle execution without changing preprocessing.

No description available

Optimization framework for large-scale MCP service integration with hierarchical routing, RL-based tool selection, and parallel execution

A human-defined rule-based trading strategy (directional anchor) , reinforcement learning (RL) agent that optimizes execution decisions with strong risk management, monitoring, and validation layers

A full-stack Auto-ML platform integrating XGBoost, Prophet, and Deep RL. Features a professional Streamlit dashboard, CLI, and optimized execution pipelines for tabular data analysis

A real-world OpenEnv RL environment where an AI agent acts as a Database Administrator (DBA) to optimize slow SQL queries by managing indexes and reducing execution cost.

An experimental, small-scale RL framework designed to profile and optimize the execution latency of verifiable reward functions. This project implements a RL training loop for a 1.5B parameter coding model, specifically targeting the "Evaluation Bottleneck" where standard containerization (Docker) throttling limits GPU utilization.

**RCRacer** is a modular, deterministic racing simulation framework for controller development, RL training, and evolutionary optimization. It features a strict layered architecture separating simulation, agents, execution, and GUI—ensuring reproducibility, scalability, and clean experimentation.

Built a real-world task scheduling environment using OpenEnv standards. Implemented a Q-learning agent to optimize CPU job execution based on priority and duration. Deployed an interactive RL dashboard using Gradio on Hugging Face Spaces for live simulation and evaluation.

SmartFlow enhances bike-sharing efficiency by combining deep reinforcement learning with agentic AI. The RL model optimizes bike distribution, while agentic AI coordinates real-time actions, like alerting truck drivers. This scalable approach ensures smart decisions and timely execution for urban transport.