I’m a 3rd-year PhD student advised by Prof. Michael Cafarella in the Data Systems Group at MIT CSAIL. I previously worked with Prof. Rachel Pottinger and Prof. Raymond Ng in the Data Management and Mining Lab at the University of British Columbia.

Today, with the explosion of data, more and more people are in desperate need to access and make use of data in more and more fields, such as social science and healthcare. However, both domain-specific and programming skills are required for an individual to do so, but not everyone has the background. Inspired by such use cases, my current research focuses on developing methods to help users interact with and make sense of data using machine learning and programming language techniques.

On-Going Projects

Bottom-Up Standardization for Data Preparation Programs

Data preparation is the process of cleaning, transforming, and structuring raw data into a format that is usable for downstream data-powered applications. Many data pipelines require a custom data preprocessing program. Although these programs embody technical and domain insights, they are difficult for others to understand and reuse because they are pipeline-specific and are written in general-purpose languages. As a result, these programs can easily become a breeding ground for poor engineering and statistical practices. Ideally, preprocessing scripts would be as standardized as possible among applications of a database, with a minimal portion tailored to the particular data application.

We propose a novel script-standardization system, in which users start with a traditional application-specific draft of a preprocessing script. The system finds a standardized version of the user script that is “admirably boring” – one with a fraction that is more common — and thus more widely-tested and more easily-understood — while preserving the user intent in the original script. We present an algorithmic framework that yields such scripts and implemented a prototype system, LucidScript. We evaluate our approach against several state-of-the-art methods, including GPT-4, on six real-world datasets, showing the effectiveness and robustness of our approach by improving script standardization up to 42% w.r.t. the user intent.

Publications

Workload-Aware Query Recommendation Using Deep Learning To Appear in IEEE EDBT ‘23.
Eugenie Y Lai, Zainab Zolaktaf, Mostafa Milani, Omar AlOmeir, Jianhao Cao, Rachel Pottinger

Summarizing Provenance of Aggregation Query Results in Relational Databases IEEE ICDE ‘21: 1955-1960. Omar AlOmeir, Eugenie Lai, Mostafa Milani, and Rachel Pottinger.

Pastwatch: On the Usability of Provenance Data in Relational Databases IEEE ICDE ‘20: 1882-1885.
Omar AlOmeir, Eugenie Lai, Mostafa Milani, and Rachel Pottinger.

Other

Blog for fun.

Miscellaneous to de-stress.