Statement of purpose for master's and PhD programs: 8 examples

At 11:40 on a Friday night in March 2025, the primary database behind our payments service in Pune failed over to a replica that was four seconds behind. In those four seconds, 312 customers paid for their train tickets twice. I was the engineer on call. By Monday the refunds were done and I had written the postmortem, and I had realized something uncomfortable: I could describe exactly what went wrong without being able to say why our design allowed it. That gap is why I am applying for the MS in Computer Science at [University], with a focus on distributed systems.

I studied computer engineering at [College], where my strongest courses were operating systems and computer networks. My final-year project was a small key-value store that copied every write to three Raspberry Pi boards. It worked until I unplugged one at the wrong moment, which taught me more than the parts that worked. My transcript has no course in distributed algorithms, because my college did not offer one, so I have tried to fill the gap myself. Over the past year I worked through the Raft paper and an online distributed systems course, and I wrote a toy Raft implementation in Go that passes most of the course's tests and fails the ones about network partitions in ways I still don't fully understand.

For three years I have been a backend engineer on the payments team at a fintech company that processes about 1.2 million transactions a day. I own the idempotency layer, the code that is supposed to make sure a retried payment is charged once. After the March incident I led the redesign: we moved from timestamp-based deduplication to idempotency keys stored with each transaction, and added a check that refuses writes when a replica is behind. Duplicate charges in the next six months fell from a few hundred a quarter to two, both traced to a bank's API rather than ours. The work made me better at reading other teams' failure reports, and it convinced me that the hardest bugs in our system live at the boundaries between services, where no single team's tests reach.

That is the problem I want to study. At [University], the distributed systems sequence ([course numbers]) would give me the foundations I lack, and [Professor Name]'s work on testing distributed systems by injecting faults is the closest thing I have found to a way of catching boundary bugs before customers do. I would like to take the thesis option and work on a project in that area, ideally one that tests real open-source databases. I also follow the department's systems reading group, which publishes its paper list; I have read eleven of this term's papers so far, and the discussion of [paper] changed how I think about retries.

After the degree I want to return to building payment infrastructure, as the engineer who designs for failure rather than the one who explains it afterward. Millions of people in India now pay for groceries, rent and train tickets through systems like ours, and I would like to spend my career making those systems boring in the best sense: correct when something breaks.

Last month our team ran a planned failover test in the middle of a working day, with all of us watching. Nobody was charged twice. I would like to understand why it worked as well as I understand why it once didn't.

For four years I have worked in revenue cycle at a regional hospital network, where my job is to explain why insurers deny claims. Last year I noticed that denials for one outpatient procedure doubled every January, and I spent a month in spreadsheets proving the pattern was real before a director asked the question I couldn't answer: was it a pattern, or a bad month repeated? I didn't know how to tell. Since then I have completed Calculus II and Linear Algebra at [Community College] in the evenings, with an A in each, to meet the program's prerequisites, and I have taught myself enough R to fit the regression my director wanted, without being confident that I fit it correctly. The MS in Applied Statistics at [University] would give me the methods behind the tools I already use, and its consulting practicum is the closest thing to my current job that comes with a statistician checking my work.

For two years I have volunteered at the [County] Historical Society, cataloging a collection nobody had opened since it was donated in 1988: the meeting minutes of a cannery workers' union local in Astoria, Oregon, from 1934 to 1941. Most of the minutes are about wages and the length of the season. But from 1938 the secretary began recording, in the margins, which members spoke Finnish, Chinese or English at each meeting, and how motions were translated. I want to know why she started, and what translation did to who could speak at all. My undergraduate thesis at [College] was on labor newspapers, so I know how to read union sources against the grain, but I have never worked in more than one language or written about migration. The thesis track, [Professor Name]'s work on immigrant labor in the Pacific Northwest and the oral history seminar would give me what I need to turn a box of minutes into an argument.

I have been a case manager at a family shelter in Milwaukee for three years, and the question I ask most is not about housing. It is “Where do the kids go to school?” Children who move into the shelter in the middle of a semester can miss weeks before anyone enrolls them, and I have learned the federal and district rules well enough to get most of ours back in class within four days. What I can't do is help the eleven-year-old who stops talking once she is there. I want clinical training, and I want to use it where those children spend their days. The MSW at [University], with its school social work concentration and field placements in [District] schools, is built for that, and its course on trauma-informed practice with children answers the gap I feel most often at work. After graduating I plan to work as a school social worker in a district with many students who have no stable home.

For nine years I have been a court reporter in Cleveland, where my job is to write down every word and none of my own. I have typed other people's worst days at 225 words a minute: eviction hearings, traffic pleas, a man explaining to a judge that his dog was the only witness. At night I write fiction about the same rooms. My collection in progress, Let the Record Reflect, is nine linked stories told by people who pass through one courtroom in a single year; the writing sample is the first and the seventh. I am applying to the MFA at [University] for three things I can't get on my own: two years of time, a workshop that will tell me where the stories go slack, and the chance to teach. [Writer Name]'s fiction, built almost entirely from overheard speech, is the closest thing to what I am trying to do, and the teaching fellowships would let me find out whether I can teach a room of first-year students to listen the way my job taught me to.

For the past two years I have coordinated a lab at [University] that studies how toddlers learn words from screens. I ran 140 of the 212 sessions in our study comparing live video chat with prerecorded video, and I noticed something the design didn't measure: on video chat, the children who learned the new words were mostly the ones whose experimenter paused when they pointed. With [Professor Name]'s supervision I coded the session videos for those pauses as an independent project. My first coding scheme failed a reliability check between two raters, so I rewrote the definitions and recoded everything. Children who got more contingent pauses learned more words, but with 31 children I treat that as a hypothesis, not a finding. I presented it as a poster at [conference] and wrote it up as my honors thesis.

I want to find out which part of contingency matters for word learning: the timing of an adult's response, or the shared attention it signals. [Professor Name]'s experiments that separate gaze from timing in infant attention, and [Professor Name]'s work on word learning in bilingual homes, are the two lines of research I would most like to join, and the program's training in longitudinal methods would let me ask the question over months rather than in one session.

Nineteenth-century story papers, the cheap weeklies that serialized fiction for working readers, printed their readers' letters, and the letters argued with the fiction. In one 1872 run I read for my master's thesis, a reader complained that a heroine had been allowed to marry above her station, and three weeks later the serial gave her a dowry from a long-lost uncle. My thesis at [University] traced twelve such exchanges and argued that the papers treated serial fiction as a conversation with their audience rather than a finished product. It left me with a question it couldn't answer: how often did readers change the stories, and whose letters did editors choose to print?

To answer it I need to read across hundreds of issues, not twelve, and to learn book history and computational text analysis, which my master's program didn't offer. [Professor Name]'s work on nineteenth-century reading publics and [Library]'s holdings of [story papers] make [University] the place to do it, and the department's certificate in digital humanities would give me the training to read at that scale.

In two years as a process engineer at a solid-state battery startup, I made about 3,000 ceramic electrolyte pellets, and I learned to predict which ones would crack. Pellets sintered at the edge of the furnace tray failed after a few hundred charging cycles far more often than pellets from the center, and the cross-sections I imaged under a scanning electron microscope showed cracks starting at grain boundaries where the grains were largest. We fixed the problem the way companies do: we stopped using the edge of the tray. I want to understand it instead. Undergraduate research in [Professor Name]'s ceramics lab at [College] taught me sintering and X-ray diffraction, and the startup taught me to design experiments that answer a production question in a week. What I lack is the mechanics, and the in-situ methods to watch a crack form while a cell is cycling. [Professor Name]'s group does exactly that, and [University]'s [facility] would let me run the experiments. After the PhD I plan to lead materials research in industry.

My first-year chemistry grades (a C and a C+) come from a year in which I worked 30 hours a week to cover my family's rent. I retook general chemistry in my third year and earned an A, and my GPA in the major since then is 3.8.