I built an AI-powered natural-language query system at Oil & Natural Gas Corporation (ONGC) that retrieves information from structured databases and unstructured documents across 50+ document chunks.
I developed semantic retrieval pipelines with FAISS and Sentence Transformers, improving contextual retrieval efficiency by approximately 28%. I also engineered a RAG question-answering system with Ollama and Mistral that reduced irrelevant responses by nearly 30%.
For a healthcare project, I developed an ensemble deep-learning system using EfficientNet, ResNet50, TensorFlow, and 5,000+ dermoscopic images to classify skin lesions with approximately 85% accuracy. I added Grad-CAM explainability and a Streamlit application for image uploads, prediction visualizations, and recommendations.
I also built C-Lite, a Mini C compiler using Flex, Bison, Python Tkinter, and GCC, with lexical, syntactic, and semantic analysis, a hash-based symbol table, and Three-Address Code generation.
