Hi, I'm Alan!

I work on applied machine learning solutions for Plaid, and am currently focused on reducing fraud in our flagship Remember Me product. Previously, I built and launched the Retention Score, an ML-trained model that predicts if a user will switch their direct deposit from one bank to another.

Highlights

• Productionized Retention Score, an institution-level model for identifying direct deposit churn risk; designed Balance and Route Hits features contributing 16.39% and 9.73% total feature gain respectively.
• Engineered a transaction-profile feature pipeline over ~800GB of training data, preventing EMR OOM failures and improving Remember Me fraud recall from 59.51% to 63.60% at a fixed 10% block rate.
• Owned the Go backend for the Connection Risk Score MVP, exposing link-time trust levels to banks with online scoring, feature fetching, database writes, DLQ handling, integration tests, and Grafana monitoring.

Recognition

The following is copied from my 2025 Performance Review

• "First: congratulations on the 4 rating. Genuinely. You're the only new grad on the product eng[ineering org] who earned that in their first review cycle, and that reflects real impact in a high-stakes environment. That's exceptional, and I want you to know I see it clearly." - Direct Manager
• "In [the] first ~6 months as a new-grad SWE, [Alan] stepped into a missing ML engineering gap during a high-visibility launch and demonstrated rapid technical growth on complex problems." - Direct Manager
• "Alan consistently stepped outside of his core role as a Software Engineer to support the development of the Retention Score. [H]e took on responsibilities well beyond his immediate scope, approach[ing] this work with a strong sense of ownership and curiosity, [and] often operating with the mindset and accountability of a more senior teammate." - Colleague
• "Develop[ed] an MVP retention score by teaching himself the necessary modeling techniques and building the solution with minimal support, enabl[ing] beta test[s] with multiple data partners on a tight timeline. Informed and managed key tradeoff decisions around the model." - Colleague

To see more thoughts about the internship, please visit the blog I wrote with Plaid and my LinkedIn Post.

Technical Details

• Built decryption pipeline for 200 million rows, unblocked top company-wide project, and brought sprint back on schedule.
• Constructed database with Apache Spark and Airflow to query sensitive AWS lakehouse resulting in 7 times faster lookup.
• Completed capstone project, including stretch goals, three weeks early; received internship extension for second capstone.

Overall Internship Rating

I was evaluated (out of four) on the following categories: Scope, Technical Skills, and Leadership. I received a 4/4 in all categories, a 4/4 overall, and a return offer.

Comments from Manager's Written Performance Review

• "Alan has demonstrated exceptional ability to quickly learn and apply new technologies."
• "Despite not having formal SQL training, Alan took the initiative to learn SQL concepts and effectively applied them to his capstone project."
• "Alan excels at reaching out and working with people both within and outside of his immediate team."

Comments from Other Plaids

• "MVP of summer 2024. You brought so much joy to the team, while shipping a ton of impact." - Henish Shah
• "Congrats Alan! You shipped a ton of impact this summer; can't wait to see what you do next year!" - Samir Naik

Technical Details

• Improved testing workflow by using CAN-Ethernet to flash and update firmware, resulting in thirty times faster execution.
• Reduced excess logs by 98% as measured by average .CSV file size by implementing a Rust update on change program.
• Fixed HTTP 429 error using Actix-Web and Rust allowing internal tools to access up-to-date information without timeout.

Picture of me next to the then-new Cybertruck.

Technical Details

• Architected ML based video quality trade-off algorithm using frame complexity resulting in 30% gains in efficiency.
• Resolved critical bug in ad-stitching for all video output by developing merging logic with 95% test coverage using Jest.
• Created debugging tools in NodeJS using Docker, Kubernetes, and AWS for deployment and pre-production builds.

Recommendation

It was a great privilege to have Alan as an intern at PlutoTV in the summer of 2023. Alan quickly onboarded and was up and running with important knowledge and capability much quicker than most of our experienced hires. He is an outstanding engineer with a very bright future and we would be thrilled to welcome him back.

From Mr. James Rose, direct manager and lead software engineer at PlutoTV.

Technical Details

• Led development on new video streamer that processes raw radar data, synthesizes frames, and computes analytics.
• Developed robust image decoder with full coverage using C# image library to process raw byte data to generate Bitmaps.
• Outputted low-latency live streams with minimal overhead by stitching frames and utilizing FFMPEG to create .mp4 files.
• Deployed written feature in mission control software for Sea Guardian Drone with over 10,000 real flying hours in India.

Recommendation

"Alan was immediately given a high-visibility task directly by the vice president of our division. He took on the full task himself, delivered with time to spare, and routinely offered solutions that were better than the original plan."

From Dr. Matthew Cornick, former director at General Atomics.

Co-authored a paper with the American Chemical Society. Google Scholar

• Developed UI in Python using Tkinter allowing for fine-grain control of camera API to take images of paper-based sensor.
• Designed desktop widget using QT in C++ utilizing OpenCV library to intelligently detect particles saving 60% in storage.
• Won Best Poster Presentation from the Howard Hughes Medical Institute.
• Trained artificial intelligence and scaled neural networks with PyTorch to identify the type of pollutants using holograms.

Paper Abstract

Exposure to bio-aerosols such as pollen can lead to adverse health effects. To address this need, we presented a mobile and cost-effective label-free sensor that takes holographic images of particulate matter concentrated by a virtual impactor. The resulting system achieved 92.91% blind classification accuracy across multiple pollen types.

Published as a co-author in the American Chemical Society.

• Led 10 research interns in a multi-national study on smell, smell experiences, and technological integration.
• Used PyTorch, Python, and PCA to classify and identify 10 different smells.
• Built a three-part olfactory communication system spanning detection, communication, and synthesis in C.

Awards

• 2nd place overall
• 1st in the Best Life Hacks category
• DigiKey sponsored award

• I worked on the Swift app, including the map to locate Lockrs and the item search experience.
• The hardware used Arduino and common electronics; the kiosk used React Native and Firebase.

Devpost · GitHub project

• Created a backend that fetched data from Google Datastore to dynamically update the webpage using Java, JavaScript, and HTML.
• Led the UI design by creating the layout, cards, and maps.

Project GitHub link