At Orbitek Space, I built a time-series ML pipeline to estimate orbital debris tumble rates from ground-telescope light curves. I trained gradient-boosted models and a transformer, then validated them against laser-ranging ground truth and a second telescope archive.
At DruvStar, I designed and deployed a production agentic AI system that processed mixed-format files through asynchronous AWS SQS queues. I also built NLP evaluation frameworks and guardrails that cut false positives by 40%+ against the GCP DLP baseline.
At Accenture, I redesigned a data ingestion layer with Kafka and PySpark for large distributed datasets, reducing batch compute cost by ~25%. In my Text-To-SQL project, schema-aware prompt engineering lifted valid SQL from ~38% to ~77%.
