Work Experience
DAPHNE - Open Source Data Analysis Language
Berlin, Germany
March 2025 - August 2025
During my time abroad at Technical University Berlin, I worked on the development of a data analysis pipeline
called Daphne. Daphne is a custom domain-specific language (DSL) for large-scale data analysis.
My contributions involved extending the capabilities of the DaphneDSL to make the platform more accessible to different data analysis use cases.
I converted data structures coded in high-level languages to be compatible with DaphneDSL. As it is still a prototype language, I had to modify algorithms to work in DaphneDSL's constraints.
Whenever the language was missing a crucial data structure for an algorithm, I would extend the languages functionality by making additions to its open source repo.
Specifically, using C++, I enhanced core data structure functionality by extending list objects to support the insertion and deletion of matrices of varying sizes,
which improved the language's flexibility for complex data manipulation.
As this is a public open source project, I had to make sure that the changes I made were compatible with the existing codebase.
This included creating unit tests, writing documentation, and writing detailed descriptions for my pull requests to ensure that the features I added were not breaking any existing functionality and that other developers could easily understand the changes I made.
UC Online - Automated Frontend UI Testing
University of California Office of the President—California
June 2024 - September 2024
As a Software Developer Intern with the University of California Office of the President, I was tasked with reviving and expanding the automated front-end UI testing suite for the UC Online platform.
Working in an agile environment and using Jira to manage tasks, I developed automated regression tests with Python, Selenium, and Pytest.
These tests verified that essential functions—such as user login, course creation, and student enrollment—remained fully operational on the Salesforce-backed platform after any updates.
When I began my internship, the test suite had been non-functional for three years, accumulating significant technical debt.
My primary objective was to first resolve these foundational issues before adding new tests. I refactored the entire codebase to be more flexible and resilient, ensuring it would not break with minor, unexpected changes to the Salesforce website's HTML.
This process included optimizing the class structure with object inheritance, which not only eliminated the technical debt but also reduced the codebase by 60%.
Throughout the project, I followed a test plan created by my manager, meeting with him frequently to address constraints and modify our strategy.
A key focus of my work was ensuring the long-term maintainability of the test suite.
To achieve this, I wrote extensive documentation for the code and processes, making it easy for future developers to use, understand, and extend the test suite long after my internship concluded.
Projects
Mindful Friction — iOS Digital Well-Being Application
Technologies: SwiftUI, Swift, Apple Screen Time API (FamilyControls, DeviceActivity, ManagedSettings), ActivityKit, WidgetKit, App Groups
April 2026 – Present
- Scale & Impact: Successfully launched a private beta encompassing 15 active users and exceeding 1,000 user sessions within the first month of tracking.
- Multi-Target System Architecture: Architected the application across four distinct compilation targets to isolate responsibilities, support system extensions, securely manage sandboxed device policies, and establish low-latency background state communication.
- System-Level Shield Enforcement: Secured Apple's highly restricted Family Controls developer entitlement by adhering to strict system privacy sandboxing guidelines. Integrated FamilyControls, DeviceActivity, and ManagedSettings to securely monitor device activity and inject custom, un-bypassable system application shields.
- Background Extension & State Synchronization: Programmed a custom DeviceActivityMonitorExtension to run seamlessly in the background and instantly re-engage application shields the millisecond a token session expires. Engineered real-time state persistence across sandboxed targets using a shared App Group container, leveraging UserDefaults data serialization.
- Dynamic Island & Live Activity Engineering: Developed a glanceable, real-time tracking interface using ActivityKit and WidgetKit. Designed both expanded and compact Dynamic Island presentations that swap out anxiety-inducing numerical countdowns for a custom, reverse-filling progress circle paired with an absolute completion timestamp.
- Behavioral UX Paradigms: Implemented a mandatory 3-second "hold-to-confirm" gesture alongside a custom time-selection scroll wheel. This breaks subconscious physical muscle memory and forces the user to transition from an impulsive digital habit loop into a mindful, intentional action.
LLM Translation Paper - TU Berlin
Technologies: Python, Huggingface, CodeCarbon
2025
- Conducted a comparative analysis of 5 Large Language Models (LLMs), including Llama 3.1 and NLLB, to evaluate their performance, accuracy, and efficiency in translating 7 high- and low-resource language pairs.
- Engineered a Python testing pipeline to automate the evaluation process, measuring translation accuracy with BLEU and BERT Scores, and tracking performance metrics like response latency and energy consumption using codeCarbon.
- Demonstrated that for high-resource languages, specialized translation LLMs (NLLB) are significantly more efficient, using only a fraction of the energy while achieving comparable accuracy to larger, general-purpose models like Llama 3.1 8B.
- Identified the limitations of current models in translating low-resource languages, with the top-performing model achieving only a 10% BLEU score, and concluded that BERT Score is an unreliable metric for languages with sparse training data.
- Presented the research methodology, key findings, and conclusions to an audience of international peers and instructors.
Movie Database Web App
Technologies: Java, MySQL, RESTful APIs, JavaScript, Apache, AWS
2025
- Built full stack architecture from scratch. Set up an AWS EC2 instance, Apache Tomcat, and imported a large database of movie information to a MySQL database.
- Implemented a frontend using JavaScript, jQuery, and ajax.
- Developed an ETL pipeline to parse large XML files to augment the database.
- Comprised ~20 medium features. e.g., fully functional website that displays a catalog of thousands of movies, cart checkout backed by sessions, secure login using SHA256 hashing and sessions, full-text search and auto-complete backed by a cache, bot detection using reCAPTCHA, protection against SQL injection attacks via PreparedStatements, etc.
- Deployed Containerized version of the project on a Kubernetes cluster spanning multiple AWS instances
UC Irvine Search Engine
Technologies: Python, MongoDB, Django, HTML, CSS, JavaScript
2024
- Programmed a web crawler to analyze 200K+ UCI websites, capable of detecting crawler traps.
- Designed a MongoDB pipeline capable of inserting over 1 million documents with efficient query retrieval in seconds.
- Implemented a query search engine with page ranking algorithms, tokenization, and indexing to categorize search results.
- Planned and structured the project into actionable tasks, allowing for simultaneous development with partner.
Baseball Statistical Analysis
Technologies: Python, Pandas, PostgreSQL, Django
2024
- Developed a Python app to evaluate the cost-efficiency of MLB teams relative to player salary and performance.
- Scraped and analyzed 15,000 data points of contracts and player statistics from baseball websites with pandas.
- Created a pipeline to take in the scraped data and clean it to clear inconsistencies. e.g., traded players or players with identical names (father and son)
SF Hackathon – Application
Technologies: Javascript, React, MongoDB, Neurelo, REST API
2024
- Developed “Metro Ventures”, a mobile web scavenger hunt application, using JavaScript and React.
- Implemented backend solutions with Neurelo and MongoDB as a REST API to manage and store interactive route data.
- Led a team of four to define and visualize the application design, resulting in a functional prototype.
