Jan 2026 - May 2026

Full-stack IoT system with ESP32, FastAPI, React, and real-time MQTT

Architected a full-stack IoT system for a smart home model meant to give full lighting, temperature, and security control through a React dashboard; built with ESP32 firmware in C/C++. Implemented a rules engine with configurable threshold triggers and MQTT pub/sub for real-time sensor data (temperature, occupancy, light, door lock) with sub-second actuation latency.

Mar 2026 - Mar 2026

Privacy-first occupancy tracking via ARP scanning with a React dashboard, React Native app, and FastAPI backend

Full-stack presence-sensing platform that turns any local Wi-Fi network into an occupancy engine with no extra hardware required. The Python backend performs ARP sweeps every 30 s with Scapy (nmap fallback), records devices in PostgreSQL, and exposes a FastAPI REST API consumed by both a React + Vite admin dashboard (deployed to GitHub Pages) and a React Native + Expo mobile app with background geofencing.

Mar 2026 - Mar 2026

Visual academic citation graph (ScholarGraph) built on Neo4j, FastAPI, React Flow, and LangChain

clearCites / ScholarGraph is an open-source visual discovery tool that maps academic research as an interactive knowledge graph. Papers are Neo4j nodes; citations, author overlaps, and funding relationships are edges. LangChain + OpenAI generates plain-English summaries and evaluates paper relationships (validates / builds_on / challenges). Users can navigate the full citation pedigree of any paper and identify publicly-funded research by funding-source tags.

Mar 2026 - Mar 2026

Discord bot that polls Twitch, Twitter/X (Nitter RSS), and vtuber Reddit news on a schedule with optional OpenAI summaries

streamerStalker is a TypeScript Discord bot that keeps communities in sync across three platforms without leaving Discord. node-cron jobs poll the Twitch Helix API every minute for live-stream transitions, Nitter RSS feeds every 3 minutes for tweets, and Reddit RSS every 30 minutes for vtuber news. All tracking is per-guild and per-channel; an optional OpenAI gpt-3.5-turbo integration generates 2–3 sentence summaries of news updates.

Mar 2026 - Mar 2026

Manifest V3 Chrome & Safari extension that removes YouTube Shorts and recommendations via MutationObserver

A cross-browser Manifest V3 extension (Chrome and Safari) that keeps YouTube focused by blocking the Shorts shelf, navigation entry, and individual Short items, and hiding sidebar recommendations, end-screen cards, and homepage rows. A MutationObserver and YouTube's yt-navigate-finish event keep blocking active during SPA navigations. A popup toggle lets users enable or disable blocking without reinstalling.

Mar 2026 - Mar 2026

Ensemble ML system (XGBoost + LightGBM + LR, 103 features) predicting NCAA tournament winners with a live GitHub Pages dashboard

An ML system built for the CMU Second Annual March Madness Machine Learning Competition. A weighted ensemble of XGBoost (40%), LightGBM (40%), and Logistic Regression (20%) trained on 103 differential features per matchup, including KenPom, Barttorvik T-Rank, NET rankings, and seed history. Achieved 72.3% CV accuracy (Men) and 75.1% (Women) on real Kaggle NCAA data using walk-forward CV. A GitHub Actions pipeline retrains and deploys a live bracket dashboard to GitHub Pages.

Mar 2026 - Mar 2026

Offline voice automation assistant using Porcupine wake-word detection and Faster-Whisper STT for developer workflow control

A fully offline, always-on voice assistant for power users and developers, similar to Siri or Alexa but running entirely on local hardware. Porcupine listens for a wake word ("Jarvis") with ultra-low CPU usage, Faster-Whisper transcribes the command without internet, and a YAML-configured command dispatcher opens apps, manages windows, runs terminal commands (git, npm, pytest), and navigates project directories.

Mar 2026 - Mar 2026

Electron + FastAPI desktop overlay with a 3-layer rune recommendation engine (player history → defaults → high-elo data)

A League of Legends desktop companion that detects champion select via the LCU API and automatically recommends and imports the best rune page with one click. The recommendation engine checks the player's own win history first, falls back to curated champion/role defaults, and finally to aggregated Master+ statistics. An always-on-top Electron overlay shows the recommendation instantly without disrupting gameplay.

Mar 2026 - Mar 2026

C++ Windows host + Flutter mobile client for live screen streaming and touch-to-input control over WebRTC/H.264

phoneCtrl turns your phone into a pocket TeamViewer. The C++ host on Windows captures the display via DXGI, encodes it with FFmpeg/libx264 at up to 60 fps, and streams it over a WebRTC/H.264 data channel to a Flutter mobile client. Touch gestures on the phone are translated to Windows SendInput mouse and keyboard events, giving full remote control with DTLS-SRTP encrypted traffic.

Mar 2026 - Mar 2026

Discord music intelligence bot using Spotify audio feature vectors for cosine-similarity recommendations and Music DNA profiles

A TypeScript Discord bot that turns a server into a music ecosystem. Users import Spotify playlists and the bot stores 8 audio feature vectors (energy, danceability, valence, tempo, etc.) per track. A recommendation engine computes each user's taste vector and surfaces songs from others' playlists via cosine similarity and collaborative filtering. Commands surface Music DNA archetypes, music twins, taste distance scores, and a daily randomly-picked song.

Mar 2026 - Mar 2026

CLI that deep-scans any public GitHub repo and generates a rich terminal report, narrated audio, and optional explainer video

A pip-installable Python CLI that accepts any public GitHub repo URL (or auto-detects from the current git remote) and performs a deep tech-stack analysis: languages, frameworks, databases, CI/CD, containers, auth, messaging, and cloud SDKs. Output is a Rich terminal table and a JSON report by default. Add --audio to get TTS narration via edge-tts, and --video to get PIL-rendered slides exported to MP4. Optional Ollama LLM rewrites narration into conversational prose.

Mar 2026 - Mar 2026

Local-first shadow file system with a Node.js + Chokidar agent, PocketBase sync layer, Xterm.js web terminal, and React Native mobile client

A hybrid between git, Syncthing, and a headless CMS. filePort runs a Node.js agent that watches local folders with Chokidar, syncing changes into a PocketBase SQLite database in real time. An Xterm.js web terminal UI lets users browse and edit files from any browser (including mobile via Tailscale). Edits set a needs_sync flag and the reconciler loop applies them back to the laptop. A separate React Native component and Python Supabase poller provide an alternative mobile-to-desktop sync path.

Mar 2026 - Mar 2026

Webcam-driven hand-gesture file navigator using MediaPipe 21-point landmarks and PyAutoGUI OS actions

Hook up a webcam and navigate your file system entirely with hand gestures. repoNavTouch runs a See → Think → Act pipeline: OpenCV captures frames, MediaPipe detects 21 3-D hand landmarks per frame, a GestureRecogniser state machine classifies them into a Gesture enum (pinch, swipe-left/right, open-palm-up/down, two-fingers-up), and a FileNavigator maps each gesture to OS-level directory navigation actions via PyAutoGUI.

Mar 2026 - Mar 2026

Discord bot that algorithmically indexes server messages and uses Groq LLM (llama-3.1-8b-instant) to synthesise startup/project ideas

IdeaDB silently observes Discord conversations and runs a zero-LLM, zero-external-API metadata extraction pipeline over every message: keyword frequency analysis, tech-term regex matching, file attachment classification, URL platform recognition, and embed scraping. All data is stored in SQLite. When you run !ideas, it builds a structured context prompt from aggregated keywords and sends it to Groq's llama-3.1-8b-instant to generate actionable startup/project ideas with name, pitch, problem, audience, and tech stack.

Mar 2026 - Mar 2026

Freemium SaaS converting audio to sheet music via Demucs stem separation, MT3 transcription, and async RQ workers, Next.js + FastAPI monorepo

A B2C freemium web app (similar to Moises.ai / Chordify) where users drag-and-drop an MP3/WAV file and get back downloadable MIDI and MusicXML sheet music for every instrument. A Next.js 15 dashboard posts uploads to a FastAPI backend that validates credits, queues an RQ job to Redis, and runs a Demucs stem separator + MT3-style transcriber in a Python worker. Real-time progress is polled by a JobTracker component that animates a status bar.

Mar 2026 - Mar 2026

Always-on-top Tauri v2 + Rust desktop companion that watches your coding/typing/idle state and reacts with pixel-art animations

A tiny pixel-art companion that floats over your desktop and reacts in real time to what you're doing. Built with Rust + Tokio for the systems core and Tauri v2 for a lightweight overlay (90% less RAM than Electron). The activity monitor detects the focused window every 2 s via Windows UIA / DBus, classifying it as idle, active, typing, coding, or sleeping. Sessions are logged to SQLite (WAL) and queried by DuckDB for end-of-day habit analytics. Features 10 mini-games, multiple companion types, dialogue bubbles, petting, and a full system-tray menu.

Mar 2026 - Mar 2026

Callable JavaScript Discord bot for playing music in voice channels

A Discord bot that joins voice channels on command and streams music. Built in JavaScript with discord.js, it supports standard music-bot commands for play, pause, skip, and queue management.

Mar 2026 - Mar 2026

Lightweight Node.js module that downloads a YouTube video and converts it to MP3 via ffmpeg-static, no system FFmpeg required

A lightweight Node.js utility module (no UI) that downloads the audio stream from any YouTube URL using @distube/ytdl-core and converts it to an MP3 file via fluent-ffmpeg backed by a bundled ffmpeg-static binary. Exported as convertToMp3(url, options) returning a Promise<string> with the absolute output path. Configurable output directory, filename, and bitrate.

Mar 2026 - Mar 2026

Chrome Manifest V3 extension that auto-fills job application forms with smart field detection and custom keyword-matching rules

jobApply is a privacy-first Chrome extension that saves your profile once (locally, never sent to any server) and auto-fills job application forms on any website with one click. Smart field detection matches inputs using labels, placeholders, aria-label attributes, IDs, and surrounding text. A custom rules engine lets you define keyword-to-value rules with contains / exact / starts-with matching for edge-case fields like work authorization or salary expectations.

Feb 2026 - Mar 2026

Electron desktop companion combining League of Legends live stats via the LCU API with a 10-game mini-casino coin economy

miniGamba is a two-mode desktop app with a full dashboard and an always-on-top overlay. The dashboard provides LoL summoner lookup, live in-game scoreboard (KDA, CS, gold, objectives), and champion-select detection powered by the League Client Update (LCU) API and the Live Client Data API. The overlay adds a 10-game mini-casino (slots, blackjack, coin flip, minesweeper, scratch cards, higher-or-lower, wheel, derby, dice, poker) with a coin economy driven by in-game events, daily tasks, hourly bonuses, and passive activity detection.

Feb 2026 - Feb 2026

TartanHacks fintech app with Plaid, XRPL round-ups, and d3 treemaps

Engineered a fintech budget planner in a team of 3 at TartanHacks using with Vitest fuzz testing and Playwright E2E tests via CI/CD. Built a deterministic analysis pipeline with recurring charge detection, explorable d3-hierarchy treemap with PNG/SVG export, Plaid Link and Capital One Nessie API for real bank data, and XRPL round-up micro-savings with Testnet payments.

Jan 2026 - Jan 2026

Python CLI for intelligent Dockerfile generation with self-healing builds

Architected a containerization automation CLI in Python using LiteLLM for intelligent Dockerfile generation with a self-healing algorithm that auto-diagnoses and retries Docker API builds based on project-specific context. Logic-driven CLI using Rich and Argparse with real-time build status tracking, automated runtime stability testing, and support for distroless/Alpine multi-stage builds to minimize attack surface.

Jan 2026 - Jan 2026

Educational PyTorch re-implementation of GPT training and inference (fork of Karpathy's minGPT)

A clean, readable PyTorch re-implementation of GPT (training and inference) based on Andrej Karpathy's minGPT. The core model is ~300 lines of code: a standard Transformer decoder (masked self-attention, feed-forward, layer norm) with a BPE tokeniser matching OpenAI's GPT encoding. Includes demo notebooks for a sorting task and GPT-2 text generation, a character-level language model project, and an addition task trained from scratch.

Oct 2025 - Dec 2025

AI video/img LSB watermarking security layer using RSA-2048 and stenography

Built a dual-layer security system for a deepfake application using invisible LSB watermarking, RSA-2048 digital signatures, and a stenography module to embed cryptographic metadata into image pixels surviving JPEG compression. Developed PKI infrastructure with automated key generation, AES-256 encrypted private keys, and signature verification tools; containerized the ML pipeline with Docker for cross-platform use.

Oct 2025 - Dec 2025

Manifest V3 extension to detect, classify, and optionally block covert tracking

Architected and built a Chrome Manifest V3 extension to detect, classify, and optionally block multiple covert tracking mechanisms (1×1 pixels, navigator.sendBeacon, scripted network calls).

Sep 2025 - Sep 2025

Gamified productivity web app built in a 24-hour hackathon

Team lead on a 24-hour hackathon project: a gamified productivity web app where tasks are “bomb wires” that must be cut before detonation to reinforce accountability through task verification.

Jan 2025 - Feb 2025

Webcam + voice triggers mannequin robot, OCR pipeline, and text-to-speech

Designed a mannequin "robot" that used a webcam and voice-command triggers to capture images of books, convert them to text via an API-based OCR pipeline, and generate human-like speech from the extracted text.

Nov 2024 - Dec 2024

Full-custom 8-bit R-2R DAC with layout-level optimization

Full custom design of an 8-bit R-2R Digital-to-Analog Converter (DAC) with control over gate placement, routing, and layout-level optimization; designed and sized an Operational Transconductance Amplifier (OTA).

Oct 2024 - Nov 2024

SystemVerilog game on FPGA with VGA output

Developed a hardware-accelerated Breakout-style game in SystemVerilog, deployed to an FPGA with VGA output. Designed paddle/ball movement, brick collision detection, and screen refresh logic.

Nov 2022 - Dec 2022

Minimalist evasion game built with Python + Tkinter

Solo-built a minimalist evasion game in Python with Tkinter, implementing the main loop, keyboard controls, on-canvas rendering, collision checks, and clean state transitions.