Satyam Rijwani

I am a highly motivated Machine Learning and AI professional with a strong academic foundation in Mathematics and Computing, currently pursuing a BS–MS degree from the National Institute of Technology, Agartala. With a CGPA of 8.32, my academic journey has equipped me with a rigorous understanding of statistical modeling, optimization, and computational thinking, which I actively apply to real-world machine learning and AI systems.

.

I have hands-on experience working remotely with globally distributed AI teams in high-impact roles. As an ML Evaluation Engineer at Snorkel, I designed and implemented Dockerized terminal-based tasks with deterministic oracle solutions, ensuring strict reproducibility through pinned dependencies. I built comprehensive validation suites using pytest and CI pipelines, enforcing full behavioral coverage, anti-cheating safeguards, and compatibility with large-scale LLM evaluation frameworks. My work involved iterative task refinement, debugging agent failures, and delivering production-ready task bundles evaluated across state-of-the-art models such as GPT-5 and Claude Sonnet.

Previously, as an AI Quality Assurance Analyst at Mercor, I worked on high-rigor evaluation pipelines for complex AI problem sets. I systematically audited prompts, solutions, and multi-model outputs under structured evaluation frameworks, contributing to a measurable 30% improvement in text generation quality through A/B testing and perplexity-based analysis. I also supported production-grade NLP systems by deploying scalable LLM solutions using retrieval-augmented generation and fine-tuning techniques, while strengthening model reliability through multi-step audit pipelines.

My industry exposure began during my internship at IIT Ropar, where I worked on computer vision and deep learning research. I trained and optimized models using CNNs, ResNet50, GANs, and Transformers for image enhancement, detection, deblurring, and color correction tasks, achieving up to 93.5% accuracy. This experience strengthened my ability to experiment across architectures, loss functions, and augmentation strategies to improve both training stability and perceptual quality.

Beyond professional roles, I have built end-to-end ML systems, including an Anime Recommender System deployed using Jenkins and Kubernetes with DVC-based experiment tracking, and a deep learning–based Anime Face Generator leveraging advanced CNN architectures. My technical skill set spans Python, PyTorch, TensorFlow, Docker, cloud-native deployment, databases, and modern LLM tooling such as LangChain, RAG pipelines, and Hugging Face.