← Back

Work

Research

Robust Query Optimization in DuckDB

In Progress

Implementing robust query optimization techniques in DuckDB, starting with Predicate Transfer, as a DuckDB extension. Advised by Prof. Xiangyao Yu.

C++DuckDBQuery Optimization

SqueezeCache: Beyond 'Optimal' Eviction for Data Analytics

In Review | PVLDB 2026

Traditional caches make binary keep-or-evict decisions, but queries often need only a fraction of cached data. SqueezeCache "squeezes" entries before eviction, retaining compact representations like string prefixes or date components that can still answer common predicates. Achieves up to 4× higher cache hit ratios and 22× lower latency.

Xiangpeng Hao, Nikhil Nayak, Proteet Paul, JP Guthi, Andrew Lamb, Jacopo Tagliabue, Andrea Arpaci-Dusseau, Remzi Arpaci-Dusseau. "SqueezeCache: Beyond 'Optimal' Eviction for Data Analytics." PVLDB, 2026. (In review)

RustCachingData Analytics

Projects

Range Filters (DIVA)

Rust implementation of DIVA, the VLDB 2025 Best Research Paper winner. A range filter that achieves 1-90x speedup over Bloom filters with O(log log U) complexity for reads and writes, plus support for dynamic updates.

RustData Structures

Congee Compact

Memory-efficient compact set data structure based on Adaptive Radix Tree (ART). Achieves 4-52x memory reduction and 10-50% performance improvement over standard ART.

RustARTData Structures

SuRF

Rust implementation of Succinct Range Filter (SuRF), based on the SIGMOD'18 Best Paper "SuRF: Practical Range Query Filtering with Fast Succinct Tries".

RustData Structures

Experience

Microsoft - Azure SQL DB

Software Engineer 2·Nov 2021 - Aug 2024

Worked on transaction recovery and log redo for Azure SQL Hyperscale read replicas. Built auto-mitigation systems for network and storage incidents. Drove the development of performance benchmarking frameworks for Azure SQL offerings.

C++C#

Akamai Technologies

Software Engineer·Jul 2019 - Oct 2021

Built high-throughput data pipelines processing terabytes of packet capture data daily. Designed an LSTM-based predictive maintenance model achieving 87% accuracy in forecasting server failures. Also worked on building CI/CD pipelines, network debugging tools, and monitoring infrastructure.

PythonJavaScriptElasticsearchKafka

Side Projects

Mushare

Have you ever tried sharing a song with someone but they can't open your link because they use a different platform? Mushare is a simple app that converts links between Spotify and Apple Music.

SvelteFastAPIPython

Melophiliacs

Yet another Spotify stats site. None of the existing ones had the specific stats I was looking for, so I built one that did.

FastAPIReact

Stretches.fitness

I've been to physiotherapy enough times to know a lot of good stretches, ones that helped me recover from injuries and improve flexibility. Wanted to build something that lists them all by body part. These are stretches taught to me by physiotherapists, so I can confirm they work. More to come.

Svelte