I’ve done a bunch of miscellaneous projects in recent years. I was definitely more enthusiastic for some projects compared to others, but all of them taught me some useful lesson.

TA’ing

• 15-210: PowerLab SML autograder. Private to 15-210.
• 21-259: recitation worksheets

Hackathons

They’re fun, if only to meet cool people and see interesting projects get prototyped very quickly.

as a participant

• MuFFLe at NVIDIA x CMU 2026 Federated Learning Hackathon, Most Collaborative Award
• MolSnap, St. Jude Knowledge in Data Science (KIDS) 2025 Hackathon, Most Technical Project. Private to St. Jude Children’s Research Hospital.
• MyWriter, TartanHacks 2025.
• MSDS Summarizer, NOVA 2024.
• CMU Study Group Finder, HackCMU 2024.

as an organizer

• AWAP26 Matchmaking Server. Private to ACM@CMU.
• HackCMU 2025 Judge
• CMU x IDQuantique Innovation Track, Future Leaders in Quantum (FLiQ) 2025.

Courses and Research

A summary of course final projects and research. Notably, I’ve omitted projects that are just part of the course.

• 15-442 Course Project: Mangrove, a memory-aware scheduler for optimizing DAG execution on constrained hardware.
• MolSnap Toolbox: extending MolSnap with automated literature review agents. Private to St. Jude Children’s Research Hospital.
• 11-785 Course Project: Open-Ended Self-Edits, experiments in self-improving LLMs via self-edits on training data
• 99-270: Summer research, synthetic data generation of chemical reaction diagrams
• 02-181 Course Project: Modifying DADA2
• 21-241 Course Project: PageRank vs. HITS Analysis

Ideas and WIP

As I work on course projects and general software things, I think developer tools are nice projects that seem small but really increase the quality of life. I’m glad that Ron Minsky agrees (c.f., 411 S26 guest lecture).

Auto-Formatters are Great

I think some of my most enjoyable programming experiences were using Go for the first time for AWAP 26 server development - Go’s very opiniated auto-formatter (which also alphabetizes your imports!!) was just great for reducing my cognitive burden.

I will say that the OCaml Jane Street profile for autoformatting was considerably more annoying, as it had this habit of squashing any unnecessary line breaks. So while I like Jane Street OCaml’s heavy usage of named arguments (~f and ~init for list folds for example), I really felt like writing the 411 compiler was reading just a ginormous wall of OCaml text. It doesn’t help either that OCaml forces repeated let...in... unlike SML which has let-binding blocks let...in...end. Multiple declarations in a let get annoying because then it becomes like a tuple declaration, and that’s annoying as hell to extend because if you have like 10 of those, you have to make sure you match up where you want to add a new declaration. And if declarations depend on each other it’s even more annoying.

Contracts are Frustrating

Another frustrating experience was writing Python contracts while I was doing my AP Research project. In general Python and ML-related libraries (PyTorch, NumPy) are so annoying because of the broadcasting rules and shape mismatches. Just way too much cognitive overhead to write some code.

People can’t name things

Write own programming language that is just the semantically meaningful counterpart of parallel CPP. For example, transform becomes map, and C++ map becomes dictionary as it should have always been. option should exist as a type - SML did good there.

Other cool aspirational goals

• Contribute to LLVM somehow?
• Rewrite 411 compiler, but in C++20, 23, or 26 so I can use concepts and shit

Anyways, onto my ideas.

WIP

Shelved Projects

• Jstarb, a job tracker in Zig. I threw in the towel because I would essentially just be replicating my Google Sheet, but much worse… and there aren’t any fun systems programming things since the whole thing was going to run locally as a single-threaded application.
• tpp, solely because I wanted to try out Zig instead of dealing with C++.