The work / An ongoing collection

From an idea
to something useful.

Independent experiments and professional systems. Different environments, the same curiosity about how things work.

13 projects

An agent’s recorded view of Chrono Trigger
CHRONOBENCH / FIELD STUDY 038 CHECKPOINTS

Independent exploration

ChronoBench

LLM benchmark measuring how far frontier models progress through Chrono Trigger using an autonomous, vision-based game agent. Live leaderboard with per-run telemetry, checkpoints, and cost tracking.

LLM EvalsAI AgentsBenchmarking

Round 1 map

vp_center (victory)north (factory)south (depot)obj1 (factory)obj2 (depot)obj3 (income)obj4 (factory)obj5 (depot)obj6 (income)obj7 (factory)obj8 (depot)obj9 (income)CrimsonDrift-1_scoutCrimsonDrift-2_scoutCrimsonDrift-3_scoutCrimsonDrift-4_scoutCrimsonDrift-5_scoutCrimsonDrift-6_scoutCrimsonDrift-7_battleCrimsonDrift-8_battleCrimsonDrift-9_battleCrimsonDrift-10_battleCrimsonDrift-11_battleCrimsonDrift-12_battleCrimsonDrift-13_heavyCrimsonDrift-14_heavyCrimsonDrift-15_heavyCrimsonDrift-16_heavyCrimsonDrift-17_heavyIronSentinel-1_scoutIronSentinel-2_scoutIronSentinel-3_scoutIronSentinel-4_scoutIronSentinel-5_scoutIronSentinel-6_scoutIronSentinel-7_battleIronSentinel-8_battleIronSentinel-9_battleIronSentinel-10_battleIronSentinel-11_battleIronSentinel-12_battleIronSentinel-13_battleIronSentinel-14_battleIronSentinel-15_battleIronSentinel-16_battleIronSentinel-17_heavyIronSentinel-18_heavyNullRider-1_heavyNullRider-2_heavyNullRider-3_heavyNullRider-4_heavyNullRider-5_heavyNullRider-6_heavyNullRider-7_heavyNullRider-8_heavyNullRider-9_heavyNullRider-10_heavyVanguardStorm-1_scoutVanguardStorm-2_scoutVanguardStorm-3_scoutVanguardStorm-4_scoutVanguardStorm-5_scoutVanguardStorm-6_scoutVanguardStorm-7_scoutVanguardStorm-8_scoutVanguardStorm-9_scoutVanguardStorm-10_scoutVanguardStorm-11_battleVanguardStorm-12_battleVanguardStorm-13_battleVanguardStorm-14_battleVanguardStorm-15_battleVanguardStorm-16_battleVanguardStorm-17_battleVanguardStorm-18_battleVanguardStorm-19_heavyVanguardStorm-20_heavy
victory income factory depot
LM BATTLE ARENA / SEASON 01TACTICAL STUDY

Independent exploration

LM Battle Arena

LLM commanders draft tank squads, fight multi-round domination matches, and evolve between rounds — self-chosen identities, coach-driven edits, and opponent intel. ELO leaderboard and full match telemetry.

LLMMulti-AgentGame AI

Independent exploration

Ludic_Director

Python library for threaded "director" layers that enqueue LLM tool actions; host runs validation on the sim thread. No game engine dependency — tool calling, memory, persona markdown, multi-layer architecture.

PythonLLMAI AgentsGame AI

Independent exploration

GameAgentHarness

Framework for building and testing AI agents in game environments. Supports multi-agent orchestration and evaluation harnesses.

PythonAI AgentsGame AI

Independent exploration

ClassicControl

Classic control and reinforcement learning experiments. Benchmarking agent architectures against standard control problems.

PythonReinforcement LearningAI

Independent exploration

Loreofthelands

AI-driven game project exploring procedural narrative generation and world-building with LLM-powered systems.

PythonGame DevAI

Independent exploration

scorched_earth

Game project built with AI agent integration — combining classic gameplay mechanics with modern AI techniques.

PythonGame DevAI

Independent exploration

GameResearch

Research workspace for game AI experimentation — prototyping agent behaviors, decision systems, and simulation architectures.

PythonResearchGame AI
SOURCEEXTRACTVALIDATESTRUCTURE

Professional / FranData

Document Intelligence Pipeline

Multi-stage AI pipeline that converts dense franchise disclosure documents into structured, machine-readable data at scale.

PythonDocument AIData PipelineAI

Professional / FranData

AI Search & Validation System

Research tooling that surfaces findings across a large disclosure-document corpus, pairing retrieval with an AI validation layer and analyst-ready outputs.

PythonNLPAI Validation

Professional / FranData

Entity & Signal Extraction

Automated AI extraction and classification of company, franchisee, and technology signals from documents and public sources, feeding the franchise intelligence platform.

PythonAIData ExtractionETL

Professional / FranData

Data Ingestion & Scoring Automation

Automation that ingests analyst-completed research and proprietary scoring data into the central data platform, validating everything on the way in.

PythonAutomationData Validation

Professional / FranData

Core Data Platform Tooling

Internal query, upload, and operations utilities supporting the franchise intelligence data platform.

PythonDatabaseInternal Tools