1. General Purpose

To architect a digital presence that reflects real engineering capability — not a drag-and-drop template. The site runs on Next.js App Router with fully static generation, scores 98/100 on Lighthouse desktop, and serves 4 project case studies plus a technical blog with full LaTeX/KaTeX math rendering. The goal was to master modern routing, component architecture, and performance optimization at a production level.

2. My Role

Full-stack developer and designer. Engineered the entire frontend architecture with strict TypeScript type safety. Built a markdown-based content system using gray-matter, eliminating the need for a CMS or database. Implemented dynamic [slug] routing for blog posts, optimized image loading and bundle size for sub-1s load times, and designed a minimalist UI prioritizing readability across desktop and mobile.

3. Project Link

1. General Purpose

To answer a focused product question without writing a single line of game code: where does Circle Panic fit in the mobile party-game market, and what would it cost to build? The project analyzes 5 competitors across 8 dimensions (core loop, monetization, virality, target audience, weaknesses), maps them on a 2-axis perceptual chart, and builds a 60+ formula financial model comparing a 9-month, 10-FTE studio scenario ($1.29M) against an 18-month, 2-student bootstrap ($3,000 out-of-pocket). Designed as a PM-style deliverable: opinionated, sourced, and decision-ready.

2. My Role

Sole researcher and analyst. Built the competitive matrix comparing Heads Up!, Spaceteam, Truth or Dare, King of Booze, and Never Have I Ever. Designed the 2-axis perceptual map (online↔offline × static↔dynamic content) that visualizes the unoccupied quadrant Circle Panic targets. Constructed the bottoms-up cost model with 60+ live formulas covering personnel, infrastructure, marketing, and overhead — surfacing a 349× out-of-pocket cost gap and a 10.7× gap including opportunity cost. Delivered a 2-sheet Excel workbook and a dedicated portfolio case study page.

3. Project Link

1. General Purpose

To solve the speed-vs-stability trade-off in deep learning optimization. NAG converges fast but oscillates violently in noisy mini-batch environments. SGD is stable but painfully slow. AGGOA introduces a dynamic transition coefficient that monitors cost oscillation in real-time — using NAG's momentum when training is smooth, switching to SGD's stability when noise spikes. The algorithm is tested on CNNs with 1,000,000+ trainable parameters using the CIFAR-10 dataset (60,000 images, 10 classes).

2. My Role

Project lead and core researcher. Formulated the mathematical foundation of AGGOA — defining the dynamic transition coefficient based on gradient magnitude and cost oscillation. Prototyped the algorithm in Python with PyTorch and ran comparative experiments (AGGOA vs. Adam vs. NAG) on Convolutional Neural Networks. Currently analyzing convergence curves across 3 optimizer configurations to quantify improvements in training stability and speed.

1. General Purpose

To build a performant mobile application with a unified codebase serving both iOS and Android, handling asynchronous API calls and complex state management. The project was ultimately cancelled due to 2 factors: a measurable decline in target audience demand and the inability of 3 project partners to maintain coordination. The failure became the most valuable output — proving that engineering success requires both solid code and strong business alignment.

2. My Role

Lead mobile architect. Designed the component hierarchy using React Native and NativeWind, integrated RESTful APIs with async data fetching, and managed application state across screens. Despite delivering a functional prototype, I made the decision to halt the project when market signals and partner availability made continuation unsustainable — prioritizing honest assessment over sunk-cost continuation.