01 About

Charles Benello

I'm a software engineer and systems researcher at the University of Chicago. I recently finished a Software Development Engineer internship at Amazon Web Services in Palo Alto, working on Amazon Redshift's data-sharing query path, and I'll be joining Chicago Trading Company as a Software Engineering Intern in summer 2027. My background spans database internals, high-performance computing, and full-stack development.

I design and implement large-scale systems - from parallel sorting algorithms in Rust to GPU-accelerated query processing in CUDA to production AI pipelines in Python. I've published in VLDB and CIDR, working with Professor Aaron Elmore at UChicago's ChiData Lab and Professor Anastasia Ailamaki at EPFL's DIAS Lab.

GitHub LinkedIn

Education

University of Chicago

Master's degree / Financial Mathematics

Jun 2025 - Dec 2027

University of Chicago

BS / Mathematics and Computer Science (Systems specialization); Minor in Korean

Sep 2021 - Jun 2025

Dean's List

St Paul's School

2015 - 2021

Mathematics: A*; Further Mathematics: A*; Computer Science: D1

Current Work

Incoming Software Engineering Intern at Chicago Trading Company (summer 2027)
Research: a controlled study of LLM-based SQL query rewriting (under review)
Research: adaptive sort configuration across disaggregated and cloud-native hardware
Research: GPU-accelerated compression for analytical databases
Evaluating policy compliance gaps in Rust LLM code generation benchmarks
CovTSP: near-optimal transit coverage solver using real GTFS network data

Experience

Software Development Engineer Intern

AWS Redshift·Aug 2026 - Nov 2026

Worked on Amazon Redshift data sharing in Palo Alto: the path a query takes when it reads tables shared from another cluster.

Redesigned shared-table metadata linking to run after workload-management (WLM) admission instead of at parse time, so the WLM slot count caps concurrent linking rather than the query arrival rate; delivered as a 9-change C/C++ stack behind config flags
Cut median planning time by more than 10x in a 200-stream TPC-DS concurrency benchmark by removing the planner's dependency on linked metadata, using version-stamped row counts published by the producer with a safe-plan fallback when counts are stale
Designed a link-once follow-up so queries routed to concurrency-scaling clusters or retried after eviction are not linked twice, cutting median client latency by roughly a third to a half in benchmarks, and measured where it regresses so rollout can be scoped
Fixed correctness defects the new ordering exposed and validated on over 60,000 regression tests, a 185,000-query cross-cluster comparison with zero result differences, 228 new or extended integration tests, and a 155-test unit suite
Co-authored a controlled study of LLM-based SQL query rewriters (under review): extended and ran the evaluation harness on PostgreSQL (TPC-DS, SQLStorm), showing that execution-based selection on a small sample beats LLM self-judgment for every model

Skills

Query OptimizationDistributed SystemsConcurrencyPerformance EngineeringC++LLM Evaluation

Papers

An Experimental Study of LLM-Based Query Rewriting - Under review

Research + Publications

Research Focus

Large-scale data processing
Database internals
High-performance systems
GPU-accelerated compression

Research Highlights

Predictive modelling for CrocSort configuration on disaggregated and cloud-native hardware - finding diminishing returns in thread counts and near-optimal config selection without brute-force sweeps
Resource-adaptive query execution with paged memory management
GPU-accelerated compression: decompression strategies, workload modelling, and cost-based trade-offs
Policy gap in Rust code generation benchmarks: showing that code passing SWE-bench tests can still violate project-defined safety and style policies
Exploring external sort on GPUs, disaggregated hardware, and cloud-native environments - understanding how storage/compute separation and elastic resources reshape sorting performance

Advisors + Labs