OpenClaw Voice

Low-latency Fish Audio speech provider for OpenClaw, using the WebSocket TTS Live endpoint with HTTP fallback. Optimized for real-time voice channel conversation in Discord.

Discord TTS

Discord bot that speaks user-provided text in voice channels via Fish Audio TTS Live (streaming WebSocket, MessagePack).

EEG Emotion AI

A hybrid deep learning model combining 1D-CNN and LSTM networks for EEG-based emotion recognition using the DEAP dataset. Achieves 96.73% accuracy in four-class emotion classification directly from raw EEG signals, showcasing strong subject-independent performance without manual feature extraction.

OpenClaw Agent

Run a private AI assistant and a public Discord bot on one self-hosted OpenClaw gateway — multi-agent routing, custom Docker image, and Claude Code workflow for AI-assisted server management.

GPT-3.5 Fine-Tuning

This project explores the performance of fine-tuning OpenAI’s GPT-3.5 Turbo for stock price forecasting. Experiments on Bigdata22 and CIKM18 show that fine-tuning improves model's accuracy and F1-score. However, performance still lags behind traditional models, highlighting the need for further domain-specific tuning in financial NLP.

Agent Harness

A template for building AI developer harnesses — structured skills, conventions, and memory files that guide agents to work consistently and safely within a project.

Deepfake Detection

Multi-region deepfake detection fusing specialized CNN and CNN–ViT models over eyes, nose, mouth, and full face. Built on FaceForensics++ with adaptive-weight and majority-voting fusion strategies.

ArtistViz

ArtistViz is an interactive data visualization tool that analyzes lyrics and streaming data to uncover emotional trends, lyrical patterns, and mood distributions in an artist’s music. Powered by NLP and Spotify/Genius APIs, it transforms musical insights into personalized visual posters.