Projects

TimeGuard Graph-RAG

• Temporal queries like “CEO as of today” or “between 2019 and 2021”
• Graph-enhanced reasoning + versioned entity memory
• Fail-safe self-learning loop for policy updates

Open Source on GitHub

CLIP-RL

• Fine-tuned CLIP with custom RL reward design
• Improved zero-shot multimodal alignment
• Linked with research post “Zero-Shot Learning in Multimodal Systems”

Whisper RLHF

• RLHF-optimized transcription on noisy/accent-rich datasets
• Improved accuracy via policy feedback loops

Quant NLP

• Financial text scraping + sentiment model using MLP-Mixer
• Supports trading with real-time sentiment indicators