💻 Projects

Welcome to my Projects page!
This is where I showcase things I’ve built or contributed to — ranging from hands-on research to real-world AI systems. I enjoy working on challenges at the intersection of ML, product, and research.

🧠 Memorization in LLMs (ASPT-ER)

🎓 Co-Author

Developed a novel prompt-tuning strategy — ASPT-ER — to reduce memorization in large language models while preserving reasoning capabilities.

✂️ 30% reduction in memorized content
🔄 Retained in-context learning performance
📊 Benchmarked on perplexity and degradation metrics
⚙️ Tools: PyTorch, Hugging Face Transformers, Unsloth

🧪 Key Highlights:

• Engineered a dynamic prompt intervention pipeline optimized for privacy-sensitive datasets.
• Benchmarked multiple LLM configurations using perplexity ratios and unlearning degradation curves.
• Outperformed several state-of-the-art baselines in both scalability and response diversity.

More Projects on Github