Engineered a Python-based data translation pipeline that supports 10+ Indic languages, handles 800+ concurrent requests, and uses a 100,000-entry LRU cache to efficiently transform large-scale conversational data.

Engineered a memory-efficient Wanda pruning pipeline on a 3.2B Llama-3.2-3B model, pruning 50% (1.6B) of its weights with minimal impact on generation quality for Indic languages.

Developed an autonomous AI agent with LangChain to control automotive systems, achieving 95% command accuracy across 500+ test scenarios. The agent’s Gradio frontend application serves 200+ concurrent users, reducing manual system interaction time by 40%.

Engineered a multimodal RAG pipeline that queries 10,000+ fixture specifications to generate custom designs, resulting in 60% faster design cycles and 85% accuracy on engineering projects.