Data Engineering
Dataset Drift Detector (driftlog)
Public datasets change between releases and nobody writes it down. This writes the changelog for them.
33
Breaking changes found
PythonPandasPyArrowpytest
source private
HELLO, I'M
Shreela Sapkota Data Engineering
I build ETL pipelines and data-quality frameworks for large-scale US healthcare data, and make sure the numbers people depend on are actually right.
Associate Data Engineer @ Deerhold · open to data roles
01 · About
Hello again.
I'm drawn to data of every kind, and to using it, plus a bit of AI, to turn something complex and messy into something people can actually act on. Right now I'm doing that in a healthcare setting: validating data at scale, building ETL workflows on AWS, and treating data quality as a craft rather than a checkbox.
I care about who gets to build technology, not just what gets built. I support women in tech and believe inclusivity makes better engineering. When I'm not in a notebook or a SQL editor, I'm probably designing something, or lost in music.
I believe in learning by making, and this site, designed and built from scratch, is part of that.
02 · Experience
Where I've put data to work
-
Apr 2025 to Present
Associate Data Engineer · Deerhold Ltd.
Deerhold is a Boston-based health-tech company (offices in Kathmandu and Tokyo). Its PRIZM platform turns over 595 million US payer-provider contract records into price comparisons that hospitals, employers and insurers use to negotiate contracts and control costs.
- Automated 40+ CMS healthcare reference-data ETL pipelines (ICD-10, CPT/HCPCS, MS-DRG, NPI, fee-schedule crosswalks) in a config-driven Python and PySpark framework, cutting the full multi-million-record refresh cycle from over 24 hours to about 2, a 12x speedup. This Medicare and CMS reference data, the fee schedules, billing codes and provider records, is what PRIZM uses to benchmark prices.
- Built end-to-end pipeline automation for provider NPI, IPF-PPS and hospital-quality datasets: Selenium and config-driven source acquisition from CMS portals, S3 ingestion, PySpark transformation, then publication to Athena-queryable tables.
- Wrote automated data-quality test suites that run SQL validations on AWS Athena, covering schema, mapping and prod-vs-UAT comparison checks, with formatted Excel reports published to Google Drive. These replaced manual QA of million-row master tables.
- Researched drill-down logic across CMS fee schedules and provider registries to derive new analytics-ready datasets, defining the mapping rules and validating every derived table before release.
- Run QA on master data covering over a million US providers and nearly 10,000 hospitals, from provider address geocoding to hospital profiles, before it reaches client products.
- Work with data engineering and compliance teams to keep HIPAA-compliant handling in place across all client-facing healthcare datasets.
-
Sep 2024 to Oct 2024
Survey and Data Collection Team Member · Kathmandu Metropolitan City Project
- Collected images of inscriptions from religious sites to aid language translation and help preserve ancient history.
- Learned early on how much model training depends on comprehensive, carefully gathered data.
Education
Kathmandu University
Bachelor of Engineering in Computer Engineering
Feb 2021 to Aug 2025 | CGPA 3.67 / 4.00
- 5+
- Projects built
- 6
- Leadership roles
- 40+
- Pipelines automated
Languages
- Python
- SQL
- JavaScript
- TypeScript
- C / C++
Data & ML
- PySpark
- Pandas
- NumPy
- scikit-learn
- Matplotlib
- Seaborn
Web
- React
- Next.js
- React Native
Cloud & Tools
- AWS (S3, Athena)
- ETL Pipelines
- PostgreSQL
- MongoDB
- Git & GitHub
- Figma
03 · Projects
Things I've built
Each of these started with a question I wanted to answer. Here's what I built to answer it.
Machine Learning
Lyrics Genre Detection
Can you tell a song's genre from its words alone?
Pythonscikit-learnPandasTF-IDF
View source ↗
Data Engineering
SQL Sales Analytics
Analytical SQL on 1.55 million rows of Brazilian e-commerce data, and four optimization attempts, two of which failed.
11x faster
Best query speed-up
PostgreSQLSQLPythonWindow Functions
View source ↗
Data Engineering
HMDA Lending Analytics
Loaded 39.8 million real mortgage records into PostgreSQL, then measured and fixed the slowest queries. The worst one got 177x faster.
177x faster
Query speed-up
PostgreSQLSQLPythonTableau Public
View source ↗
04 · Community
Beyond the code
The project
Myra
A Nepali-first women's health app for period and cycle tracking. Records stay encrypted on the user's own phone rather than on a server, which matters a great deal for this subject in Nepal.
My role: data collection and the data work behind it. The two awards below are for the same project.
-
2nd Runner-Up
Hult Prize Myra
I ran the data collection and the data work behind Myra for this pitch. Placed third for design and social impact.
-
1st Place
YuwaXcel Myra
First place at YuwaXcel 2026, a startup pitching competition run in partnership with the U.S. Embassy Youth Council Nepal. Same team, same data work, a year further on.
-
Event Lead, Alumni Meet
KU IT-Meet · 2024
-
Logistics Lead
EmpowerHer · 2023
-
Marketing Team
KU HackFest · 2023
-
Treasurer
SEDS-KU · 2021-22
-
Volunteer Designer & Content Creator
Women in Data Science Kathmandu · 2021
-
Treasurer
St. Mary's Environment Club, St. Mary's High School, Jawalakhel · 2016-17
shreela@portfolio : try it
$ A real (tiny) SQL engine over my data. Type help, or run SELECT * FROM projects;
05 · Contact
Let's build something with data.
Whether it's a role, a collaboration, or a question about something I've written, my inbox is open.
Away from the keyboard
I sing, too.
Mostly cover songs, posted as I record them. If you'd like the other half of the picture, follow along, or subscribe on YouTube.