I build AI products, fast web apps, and systems that hold up under pressure.

Built an AI system that transforms long podcast videos into social-ready one-minute clips without manual editing.

• Fine-tuned ChatGPT and Gemini on a YouTube-derived dataset to improve clip selection.
• Implemented speaker detection and auto-framing to keep the active speaker centered on screen.
• Deployed a decentralized GPU-backed pipeline on Runpod to split processing into reliable stages.

Designed and optimized an automated Solana arbitrage bot for extremely latency-sensitive execution.

• Generated $300K in net profit over $12M in trading volume.
• Reduced reaction time below 100ms with globally distributed servers in New York, Frankfurt, and Tokyo.
• Stored hot data in Redis, tuned Linux servers, and customized Solana libraries for speed.

Built a web app for an accounting firm to model RRSP and TFSA contribution strategies and surface recommendations.

• Created a recommendation algorithm based on user income and deductibles.
• Designed an interactive flow with tables and visual result summaries for non-technical users.
• Delivered the product with a modern web stack and hosted backend services.

Delivered a Shopify store and an integration layer that kept catalog, inventory, and order fulfillment aligned with a supplier.

• Designed and launched the storefront, including theme customization, collections, and shipping configuration.
• Automated synchronization for more than 10,000 products using Shopify's GraphQL API and an internal database.
• Integrated supplier data so inventory and order fulfillment could be handled with minimal manual work.

Reconstructed a disassembled residual neural network from saved weight fragments and historical data, recovering the exact final permutation.

• Recovered the full 97-piece network ordering with effectively zero reconstruction error.
• Built a multi-heuristic pairing pipeline and solved one-to-one matching with the Hungarian algorithm.
• Designed greedy and simulated-annealing ordering solvers, plus visual diagnostics to guide iteration.

Reverse-engineered a massive sparse integer neural network into an interpretable hash pipeline, reducing the problem to a preimage search.

• Decompiled a 2721-layer sparse network into explicit equations and recognized its MD5-style structure.
• Built several solver paths including symbolic analysis, SAT encoding, and hash-level validators.
• Recovered the winning phrase and validated the model hypothesis against large randomized checks.

Built a local competitive AI sandbox for a CodinGame Snakebird challenge, with custom bot families, replay tooling, and benchmarking reports.

• Created a local simulation and visualization workflow for iterating on multiplayer snake bots.
• Developed A*-based planning with danger scoring, race awareness, and ally coordination ideas.
• Added benchmark harnesses and HTML reports to compare strategies beyond simple win/loss counts.

Built a modular experimentation framework for ARC-style agents, with pluggable bots, benchmark tooling, and prediction-based evaluation.

• Created a shared framework for running and comparing multiple agent architectures on the same tasks.
• Added structured board and transition summaries so agents could reason over interpretable state descriptions.
• Measured strong short-horizon prediction accuracy while documenting the gap to actual task completion.