Interests: DeFi; building with LLMs; systems that learn via evolution and adaptation with applications to algorithmic trading and agent-based decision systems.
Oracle-Anchored LVR Recapture with Selective Dutch Auctions
TL;DR. AMM design for recapturing loss-versus-rebalancing by anchoring pool prices to an oracle and selectively routing arbitrage through Dutch auctions, improving LP outcomes while preserving efficient price discovery.
Evolving Deep Learning Optimizers
TL;DR. Evolutionary framework that discovers novel deep learning optimizers by combining primitive gradient operators; produces hybrid update rules that outperform AdamW in low-data regimes.
Genetic Algorithm for Optimal User Order Routing
TL;DR. Multi-objective evolutionary solver (NSGA-II hybrid) for CoW Protocol batch auctions that balances user surplus, gas, slippage and risk under tight deadlines; outperforms deterministic baselines on small-medium orders and converges within ~0.5s median.
-
Alpha Router β Evolutionary solver
Implementation of my published hybrid genetic algorithm (NSGA-II + deterministic refinement) for CoW Protocol batch auctions. Handles multi-objective optimization, including surplus, gas, slippage, and risk, under tight latency constraints; deployed as a production-ready routing engine.
Repo: https://github.com/mmarfinetz/Alpha-Router -
evo-optimizer β Evolutionary optimizer discovery for ML
PyTorch framework that evolves new update rules by combining primitive gradient operators, including sign-grad, momentum, and normalization terms. Produces hybrid optimizers that outperform AdamW in low-data regimes and accelerate convergence across CIFAR-10, MNIST, and Fashion-MNIST.
Repo: https://github.com/mmarfinetz/evo-optimizer -
VinRouge β On-chain trading agent
Advanced signals, strategies, and recommendations, including mean reversion, RSI, Bollinger %B, and Z-score.
Repo: https://github.com/mmarfinetz/VinRouge -
Arbitrage Bot
TypeScript/Hardhat bot that discovers and executes arbitrage via Flashbots, including regular, MEV-Share, WebSocket, and hybrid modes. React dashboard, gas logic, and circuit-breaker protections.
Repo: https://github.com/mmarfinetz/arbitrage-bot -
ElizaOS β Agent framework contributions
PRs: #2869 Β· #2784 -
Tetris AI β Genetic algorithms + Markov-chain learning for gameplay
Repo: https://github.com/mmarfinetz/Tetris
Blockchain
ML / AI
Web
Infra / Deploy
Email: mitchmar@sas.upenn.edu Β· GitHub: https://github.com/mmarfinetz Β· X/Twitter: https://twitter.com/mmarfinetz7 Β· Telegram: https://t.me/mitch0x



