Akshad Agrawal

Architected an Agentic RAG system using LangGraph to audit legal contracts. Employed a cyclic "Drafter-Critic" workflow that iteratively refines risk assessments before final output. Built a Self-Healing Knowledge Base that synthesizes defensive policies from LexGLUE datasets using LLMs.

Developed a production-grade MLOps pipeline to automate training, validation, and versioning of a T5 Transformer model using Dagster. Implemented DVC for reproducibility, MLflow for experiment tracking, and custom drift detection sensors to prevent model degradation.

Engineered a platform to screen and rank resumes using evidence-based matching. Constructed an asynchronous 5-stage processing pipeline (parsing, extraction, scoring, reasoning) using Groq API and SQLAlchemy for persistent state management.