Yash Vishe

About

Backend/ML systems engineer with experience at SDSC, UCSD, and Media.net, building production semantic dataset/model discovery using embeddings + Solr KNN and LLM-based intent/reranking at 5M+ dataset scale. Emphasizes offline/online separation for predictable serving, has delivered measurable gains (23% retrieval accuracy, 38% latency reduction) and helped secure a $3M+ NSF grant.

Experience

Education

Key Strengths

• Designed production semantic discovery backend for 5M+ datasets using offline semantic indexing + fast query-time retrieval
• Improved semantic retrieval accuracy by 23%
• Architecture tradeoff expertise: front-loaded offline compute to avoid expensive runtime LLM calls and stabilize p99 latency
• Built scalable FastAPI APIs with strict schemas, stateless handlers, async jobs, caching/pagination, and autoscaling behind load balancers
• Led incremental data pipeline refactor/migration with parallel runs, automated data quality checks, idempotent DAGs, and feature-flagged cutover
• Delivered 38% latency improvement in SQL/data infrastructure refactor (PySpark/Airflow/Redshift)
• Implemented production auth patterns: JWT, scoped permissions, and database-enforced row-level security
• Proactively identified LLM inference failure mode (long-prompt bursts causing queue buildup) and mitigated via request limits and logging
• Contributed to securing $3M+ NSF grant via production-grade system outcomes