Ankit Dash

I’m an AI/ML developer who builds LLM and agentic AI systems, including Retrieval-Augmented Generation (RAG) and multi-modal deep learning, with a strong focus on evaluation. In my internship, I designed end-to-end ML pipelines for preprocessing, feature engineering, training, and reproducible experiment documentation. I use PyTorch, LangChain, and HuggingFace Transformers to turn models into working, measurable solutions.

My flagship sleep stage detection project combined a two-level TCN approach for EEG/EOG signals with Mistral-7B to generate structured PSG reports, reaching 87% cross-dataset accuracy. I also built a hierarchical agentic malware classification system that escalates low-confidence samples and achieved 99.90% test accuracy, including per-sample risk scoring and family-level alert generation. Alongside this, I develop hybrid vision models (ResNet50 + Vision Transformer + U-Net) for pixel-level forgery segmentation, and I’m Oracle Cloud Infrastructure 2025 certified in AI Foundations.