i'm an honors CS student at penn state, and i like building software around problems i actually have — systems i can use every day rather than one-off demos.
Nytr started from one of those problems: penn state dining changes every day, but nutrition and training targets don't. i built it to combine Stacks Market availability, HealthKit data, food logs, and training history into deterministic daily guidance.
right now, i'm going deeper into LLM systems, backend infrastructure, and systems/performance engineering.
python swift javascript typescript sql fastapi postgresql pgvector node.js swiftui github actions linux git
Nytr · case study · swift, python, fastapi, postgresql · 2026
Evidence-driven decision system built around Penn State Harrisburg's Stacks Market. Combines daily dining availability with HealthKit, food logs, and Hevy training history to produce deterministic meal and training guidance in a native iOS app. 1,135 backend tests (1,124 passing in CI with Postgres) · 251 iOS XCTest cases.
Talks · case study · swift, swiftui, watchos, ios · 2026
Apple Watch meeting recorder. Durable Watch-to-iPhone file transfer, on-device transcription and Apple Foundation Models summaries, then sync to Notion. 57 XCTest cases passing in CI (iOS Simulator), plus a watchOS build.
Clage · case study · python, algorithms from scratch · 2026
Neuroevolution of Augmenting Topologies (NEAT) implemented from first principles, driving organisms in a 2D simulation world. No ML libraries (matplotlib is the only runtime dependency). 300 tests, passing in CI. Over 5 fixed seeds it solves OR and AND every time and XOR 4 of 5; sine regression is not solved yet.
Devvy · case study · javascript (node.js), discord ipc, devtools · 2026
Local daemon for coordinating Discord Rich Presence across developer tools. Speaks Discord's IPC protocol directly with arbitration, TTL expiry, heartbeats, and reconnect handling. 26 test scripts, all run in CI; releases publish only after CI passes. Released for macOS; Windows support is on main, not released yet.
ScholarAI · team repo · my pipeline · python, postgresql, pgvector · jun 2026
Government-scheme discovery platform built by a 5-person team at the USAII Global AI Hackathon 2026; our team was a finalist among 6,081+ participants. My part was the data engineering: normalization, embeddings, and PostgreSQL/pgvector ingestion for 1,008 scraped schemes.
Also: MedClarity — ask questions across medical PDFs; one Gemini request per question with page-level citations checked against the loaded documents; 66 Vitest unit tests passing in CI · 2026. PyOpt — LoRA fine-tuning of Qwen3-4B to speed up Python functions (Sep 2026). Published v1.1 run: faster rewrites 29/40 → 36/40, but 13 of the 40 test prompts leaked from training, so it is not a clean held-out result; the split is fixed in v1.2 and the model is not retrained yet.
