usama shah

At MeridianWorks Technologies, I build LLM-powered enterprise applications that let users query internal data in natural language. My RAG pipelines process more than 2 million documents, improved knowledge-retrieval accuracy by approximately 32%, and reduced AI inference costs by 28%.

I develop production AI services with Python, FastAPI, Azure OpenAI, LangChain, PostgreSQL/pgvector, Pinecone, Elasticsearch, Docker, and GitHub Actions. I’ve also built document-understanding platforms, AI agents that call internal tools, and evaluation pipelines for relevance, groundedness, hallucination, latency, and cost.

Previously at VertexPoint Technologies, I delivered machine learning systems for churn prediction, demand forecasting, fraud detection, segmentation, NLP, and recommendations. My work improved model performance by up to 21%, increased relevant recommendations by approximately 18%, and automated data pipelines handling more than 500,000 records daily.

I enjoy translating business requirements into reliable AI products, collaborating across Product, Engineering, Data, and QA, and mentoring junior engineers in ML engineering, LLM applications, testing, and deployment.