fig. 0 — a self-portrait, computed
I build machine-learning systems and the software underneath them — in C++, Python, and Java. Minors in mathematics and art & design keep the work precise, and on purpose.
currently ▌
* the curves behind this page are drawn live by a flow field — refresh for a new edition
I build models, and the systems that keep them running. The sketchbook comes along.
I'm an M.S. Computer Science student at Purdue University (B.S. CS '26, M.S. '27) working where machine learning meets systems software. My strongest languages are C++, Python, and Java: I've written an HTTP/1.1 server on raw POSIX sockets, a Unix shell with job control, and ML pipelines that run in production.
Most recently, at Envita Solutions, I built an end-to-end Python email-classification pipeline that runs as its own Docker service on Azure at 2,000+ emails a minute. I took on-call for it, wrote 180+ tests, and raised accuracy from 60% to 99.12%. Before that I built backends at FractalHive and nRoad, ML research at Purdue, and taught Java to a lecture hall of 400 as Head TA of CS180.
My minors in mathematics and art & design show up as generative graphics, like the figures on this page, which are drawn live in code.
each figure is plotted live by the algorithm named beneath it — click a drawing to re-roll its seed
Fine-tuned and benchmarked YOLO, Mask R-CNN, and RT-DETR on 9 animal classes from Open Images under Gaussian blur and JPEG compression. RT-DETR won: best baseline mAP@0.5 (0.788) and the smallest high-blur degradation (−11.2% vs −17.7% for YOLO).
A dynamic-programming hit/stay agent for the Flip 7 card game: memoized, deck-weighted expected values at every state, action by argmax. Pivoted from Q-learning after diagnosing multi-agent reward instability — the DP agent matched human win rate (50%) over 182 evaluated rounds.
A cycle-aware iOS app that uses Gemini 2.5 Flash to generate daily meal and workout plans tuned to PCOS phase, pantry stock, and real equipment. Cycle prediction via exponential-decay-weighted averaging with confidence intervals; one daily AI call with a static fallback so it works offline and inside free-tier limits.
A reading app that watches you read: real-time eye-tracking drives auto-scroll and pays out incentives per line read. React + Python over WebSockets, built in 24 hours, finalist at HackMIT.
An HTTP/1.1 server in C++ on raw POSIX sockets, with four concurrency modes: iterative, fork, thread-per-request, and thread pool. Adds CGI execution via fork/exec, HTTP Basic auth, sortable directory listings, SIGCHLD reaping, and request-timing stats.
A Unix-like shell in C with process control, I/O redirection, and pipelines built on fork/exec/wait and POSIX system calls. It has full job control (fg/bg, SIGINT, SIGTSTP) to manage concurrent child processes reliably.
A real-time live-streaming server in Docker containers, deployed publicly on a GCP VM. Automated pcap capture and packet-level analysis in Python to profile streaming performance across varied network conditions.
Trees grown from grammar: L-systems and turtle graphics in C++17 / OpenGL 3.3, with flowers and leaves generated by GLSL shaders and noise (no textures, only rules). SDL2, GLM, and ImGui drive a real-time turntable camera and live parameter tuning.
internships & research, newest first
Purdue University, West Lafayette
Machine learning · systems · artificial intelligence
operating systems — parallel computing — statistical ML — artificial intelligence — data mining — human-computer interaction
Purdue University · minors in Mathematics and Art & Design
Systems software · machine intelligence · algorithmic foundations · computer graphics
operating systems — compilers — systems programming — computer networks — security — data structures & algorithms — advanced algorithms — data engineering — computer graphics — numerical methods — probability — linear algebra
purdue cs · aug 2023 – may 2026
Six semesters with CS180 — three as undergraduate TA, one as Student Success TA, and two as Head TA. Led weekly Friday lectures and labs for 400+ students, live-coding Java and creating lecture slides; coordinated the grading team and mentored students on debugging, code optimization, and software engineering practice; as Student Success TA, provided individualized academic guidance and exam preparation.
Office hours and help sessions on asymptotic analysis, dynamic programming, graph algorithms, and proof-based reasoning; proctoring and grading for the upper-level course.
Problem-solving technique and theory support, office hours, and grading.
Open to ML engineering, software and systems engineering roles, research collaborations, and good conversations about any circle of the Venn diagram.