Aditya Arya

I’m a GenAI Engineer focused on shipping reliable, measurable machine learning systems. I’ve built end-to-end pipelines that turn messy real-world inputs into accurate decisions—backed by strong observability and production deployment practices.

At NAB Innovation Centre, I engineered an automated customer-matching pipeline using OCR-extracted name, DOB, and address. Using Databricks Mosaic AI embeddings and cosine similarity, I indexed 42 million records in 28 hours, and I built a multi-stage validation framework with weighted Levenshtein-style scoring to reach 100% match confidence.

I also added an LLM-based secondary verification layer (Claude Sonnet 4.6 few-shot prompts) with a 0.75 threshold to reduce manual operations workload by 70%. I instrumented the pipeline with MLflow and containerized it with Docker, deploying production-grade APIs via Kong API Gateway to cut end-to-end latency from over 1 minute to 14 seconds.