lakshya User

I’m an LLM generalist and evaluator focused on making large language systems more reliable. At Ethara AI, I evaluated 1,000+ LLM responses for quality, reasoning, and correctness, maintaining ~95% consistency, and I improved output consistency by 20–30% through generalization and prompt refinement.

Before that, I built ML products and pipelines end-to-end. At Celebal Technologies, I developed CI/CD pipelines with GitHub Actions and Docker for containerized ML deployments, automating AWS EC2/ECR deployment and reducing manual setup by 30% through secret handling and SSH. At Feyns Labs, I shipped an MVP ad-creative generator using GPT + image APIs and created an EV market dataset via scraping + API integration, then used KMeans segmentation.

I also enjoy turning data into measurable model gains—at Suvidha Foundation, I built a labeled headlines dataset and trained XGBoost and Logistic Regression classifiers, achieving a 20% macro F1 improvement over baseline. My projects—from grammar scoring with MFCC + spectral features to multimodal RAG with 4-bit quantization (75% less memory)—reflect the same drive: accuracy, efficiency, and practical deployment.