At Microsoft Fabric OneLake, I architected a read scale-out layer for shortcut-metadata reads that increased throughput from 12K to 120K requests per second and reduced P50 latency from 880 ms to 17 ms.
I also redesigned partition scaling and routing, and built a keyed batching executor adopted by five subsystems. For shortcut writes, I replaced pub/sub cache invalidation with read-lease invalidation to eliminate up to 30 seconds of staleness and provide read-your-writes consistency.
Earlier at Microsoft, I built backend services for shortcut transformations that process 3,000+ transforms and 7B+ rows daily, and created a durable-workflow framework adopted by multiple OneLake features. As a Software Engineer Intern on Microsoft (Azure Synapse), I automated troubleshooting-guide steps, reducing manual incident-diagnosis time by about 65%.
