Projects

OPERATOR: A Dispatch Simulator

Command a slime crew. Deploy on missions. Wait. See what survives.

Simulating a multi-agent ecosystem on high-latency, resonant planetary surfaces requires more than just a game engine—it requires a technical constitution. The challenge was building a high-performance simulation that remains perfectly synchronized across wall-clock time, regardless of whether the application is active, backgrounded, or offline.

PhantomArbiter

Solana DEX arbitrage engine with multi-strategy detection and MEV protection

Cryptocurrency arbitrage opportunities exist across Solana's decentralized exchanges (Jupiter, Raydium, Orca, Meteora), but MEV (Maximal Extractable Value) bots front-run profitable trades, capturing the arbitrage before it can be executed. Detection must happen in milliseconds across multiple DEX prices.

OpenAgent

Repo-aware CLI that generates structured directives for AI coding agents

AI coding agents need context to be effective. Generic prompts miss project-specific patterns, architecture decisions, and coding style. The challenge was creating a tool that understands your actual codebase and personalizes recommendations to your development approach.

LDLA

Local Data Logistics Agent — production-grade data factory for CSV cleaning

Data pipelines often fail on edge cases: encoding mismatches, messy headers, mixed formats, duplicates, and memory constraints. The challenge was building a production-grade local data factory that handles real-world data messiness while maintaining reliability and performance.

GreenGap

Educational ecological succession game teaching environmental concepts through mechanics

Most educational games bore players with facts. GreenGap teaches ecological succession (how landscapes recover from disruption) by embedding lessons in game mechanics rather than exposition. Players must understand ecological concepts to progress—knowledge becomes prerequisite for winning.

TurboShells

Turtle racing simulation with 20-trait genetics and sacrificial breeding

Breeding simulations often oversimplify genetics to a few traits. The challenge was implementing a 20-trait genetic system with meaningful trait interactions, while maintaining frame-independent stability and performance at simulation limits.

VoidDrift

Mining simulation set at the edge of a black hole

The asteroid fields are debris from what the black hole eats. You mine them. The station has been here longer than you. It should have been consumed. It has not been. You don't know why. Factions send bottles across the event horizon. You don't know how. Resources deduct. Upgrades apply. What happens in between is outside the frame. There is no win condition. There is no escape. The horizon is a one-way membrane.

AntSim

Evolutionary ant colony simulation with genetics and pheromone trails

Traditional ant simulations often lack genetic diversity and realistic colony dynamics. The challenge was implementing evolutionary genetics, pheromone trail systems, and queen mortality while maintaining visual clarity and performance.

Convoso Team Optimizer

Team performance analytics and optimization tools for Convoso call center managers

Convoso managers lack visibility into team-level optimization opportunities. Individual agent metrics are visible, but comparative analysis, trending, and resource allocation insights are buried in reports. The challenge was building a real-time optimization engine that identifies coaching opportunities, predicts performance changes, and recommends resource reallocation—all without leaving Convoso.

Convoso Auto-Create Lead

Automated lead creation and management within Convoso dialing workflows

Call agents manually creating leads in Convoso copy-paste data from external sources (spreadsheets, emails, third-party platforms). Each lead creation takes 30-60 seconds of manual data entry. Across a call center with 20+ agents, this represents 5+ hours of daily wasted time. The challenge was automating lead creation while maintaining data quality and respecting Convoso's workflow constraints.

Convoso Insight Lens

Real-time analytics and performance insights for Convoso dialing platform

Call center managers using Convoso have access to raw data but lack actionable real-time insights. Performance metrics are scattered across different pages, requiring manual navigation and calculation. The challenge was creating a unified analytics layer that extracts, aggregates, and visualizes Convoso's hidden metrics without disrupting the core dialing workflow.

PyPongAI

Neuroevolution research platform — evolution discovers how to play Pong

Traditional reinforcement learning requires hand-crafted reward functions and extensive training data. The challenge was demonstrating that evolution alone could discover complex behaviors without explicit rules, and showing the progression from random to skilled play.

Chrome Auto-Refresh

Intelligent automatic page refresh for monitoring dashboards and time-sensitive content

Knowledge workers (traders, analysts, managers) spend hours watching dashboards that update every few seconds. Manual refresh is inefficient and causes missed data. The challenge was creating a tool that auto-refreshes without disrupting user work, respecting rate limits, and handling dynamic content properly.

Telesero Auto-Pauser

Automatic call pause management for Telesero VoIP platform

Call center agents using Telesero spend 20-30% of their time pausing/unpausing calls between break times. Each manual pause action is a context switch that reduces focus and productivity. The challenge was creating a seamless, non-intrusive solution that respects user control while automating repetitive workflow states.

RogueAsteroid

Physics-based arcade game with procedural difficulty and modular architecture

Recreate the classic 1979 Asteroids arcade game with modern code architecture and progressive difficulty. The challenge was proving that clean architecture works even in performance-critical game code—most games are giant main loops with spaghetti logic.

rpgCore