Devanshu worked on backend data processing features and optimizations in Rust across the apache/datafusion and spiceai/datafusion repositories. He implemented a Postgres-compatible fix for the translate function, ensuring duplicate characters in the 'from' argument are handled consistently with Postgres and DuckDB, which improved cross-database reliability for ETL workloads. In apache/datafusion-sandbox, he developed predicate pushdown for the floor function, transforming floor(x) = N into efficient range queries that leverage Parquet statistics, and extended this to support decimal types in apache/datafusion. His work emphasized SQL optimization, robust unit and system-level testing, and delivered measurable improvements in query performance and data quality.