umair ayaz

• Developing GRACE, a Python SDK that implements a neuro-symbolic guardrails architecture for LLM agents, separating normative reasoning from decision-making so that safety rules can be enforced around any LLM pipeline. • Implemented the full pipeline of the SDK, spanning LLM-based grounding, symbolic reasoning over prioritized defeasible rules, interpretation, decision-making, and a guard that validates every action before execution. • Validated the architecture on 2 end-to-end demo applications, including

Research Focus: Generative AI • Conducting research on Generative AI, specifically fine-tuning Transformer architectures for scientific discovery. • Modeling complex data structures (such as molecular generation) using deep learning techniques. • Training large-scale models on High-Performance Computing (HPC) clusters using PyTorch and Slurm. Tech Stack: Python, PyTorch, Transformers, Slurm, HPC, Linux.

Core Focus: Generative AI, MLOps, and ML Pipelines • Architected end-to-end Machine Learning pipelines, integrating predictive models into production environments to drive business decision-making. • Developed RAG-based LLM applications (using OpenAI & LangChain) to optimize campaign analytics. • Automated complex analytical workflows, reducing manual reporting time by 12+ hours weekly. Tech Stack: Python, SQL, Generative AI (LLMs), Docker, Git.

Core Focus: Full-Stack Web Development • Developed production-level features for e-commerce platforms using React.js and Ruby on Rails, ensuring scalable architecture. • Engineered high-performance front-end components and managed seamless back-end integration for international clients. • Collaborated in an Agile environment (using Jira & Confluence) to deliver iterative product updates and streamline deployment cycles. Tech Stack: React.js, Ruby on Rails, JavaScript, HTML/CSS, Git.

A Levels in Mathematics, Chemistry, Physics and Computer Science