Hi, I'm Lindsey
RSS FeedI'm a backend engineer and an M.S. Computer Science student at Northeastern University. Most recently I was a Software Development Engineer Intern at Amazon Web Services, where I owned and shipped a sharing feature end to end — across two codebases that had historically diverged, and with the security review that comes with serving user-generated content to the public.
Before that I built a medallion data lake on S3 and a set of Python/FastAPI microservices. I like the unglamorous parts of the job: access-control models, idempotent pipelines, and cutting a p95 in half. I work with AI to get productive in an unfamiliar domain in days rather than weeks, then do the part it won't do for you — code humans can read, test, and audit.
I write up what I learn in the notes section, and there's a bit more context on the about page.
Featured
Does MySQL need a COMMIT after INSERT? No, and that's the problem
The answer is no, MySQL autocommits by default. The interesting part is that typing COMMIT anyway raises no error, so a wrong mental model can survive for years.
Recent Posts
Idempotent MERGE: the re-run was safe, so why did the totals change?
Merging on a natural key is supposed to make a batch job replayable. It does, right up until the source batch, the ordering, or the input window stops being what you assumed.
At-least-once delivery means your consumer will process the same event twice
SQS, Kinesis, and DynamoDB Streams all guarantee at-least-once. That guarantee is a requirement on your consumer, not a promise you can ignore.
A 90% cache hit rate and the API is still slow. Look at your connection pool.
Caching removes repeated work. Connection pooling removes per-request setup. They fix different halves of your latency, and a cache cannot touch the half it doesn't own.
Why MySQL says "Unknown column" for an alias you just defined
You aliased a column in SELECT, used it in WHERE, and got ERROR 1054. ORDER BY accepts the same alias without complaint. Here's the one rule that explains both.