Jatin Bagul

I'm building end-to-end machine learning and NLP systems as an independent, project-based Machine Learning Engineer, from preprocessing and feature engineering through training, evaluation, APIs, and full-stack deployment.

I fine-tuned Mistral-7B with QLoRA and PEFT for a hallucination-resistant LLM, using contrastive and refusal-based datasets to distinguish answerable from unanswerable queries. My work focuses on improving factual consistency through instruction tuning and faithfulness-oriented evaluation.

I built StoxMind, a full-stack stock portfolio and sentiment-analysis platform that combines NLP and ML models with interactive dashboards. I also developed a computer-vision traffic signal optimization system and a multilingual museum chatbot with visitor analytics.

I work with Python, PyTorch, Hugging Face, LangChain, RAG, and web deployment tools, and I optimize experiments under constrained hardware using an RTX 3050 and Google Colab. I'm completing a B.Tech in Computer Engineering.