Work
Projects A mix of work from my role at USLBM and software I build independently: games, persistent worlds, AI services, and developer tools. Each project page explains what I worked on and where it stands. For the smaller programs behind the projects, start with Developer tools .
01 Enterprise Document Extraction Service 02 Enterprise AI Trajectory Service 03 Procure-to-Pay Automation 04 Canonvale 05 AI-Assisted Game Development Toolchain 06 Ludex 07 Reverse Engineering Workbench 08 Galleries
Project
From identifying the need to delivering a shared document extraction service.
A shared document extraction service I conceived, designed, and built independently at USLBM.
Open the project
Conceptual overview Documents Extraction Validation Shared service Visibility Documents Extraction Validation Shared service Visibility
The same business purpose can arrive in different layouts, terms, and structures. Some follow a template. Others need interpretation.
Structured information, shared by many applications instead of rebuilt inside each one.
A flexible reading still has to be accurate and consistent enough for another system to use.
Different consumers have different requirements. The service has to absorb that without becoming a pile of exceptions.
Predictable behavior, clear interfaces, and a way to see what happened when a result fails.
Project
Identifying the need for AI execution visibility and building the shared service to address it.
A centralized AI trajectory service I conceived, designed, and built independently at USLBM.
Open the project
Conceptual overview Execution Record Correlation Investigation Account Execution Record Correlation Investigation Account
Decisions, model interactions, tool calls, and intermediate steps before a result.
A consistent account of that activity, so each application does not have to invent its own.
Records structured and related so they can be compared across a workflow.
When a request fails, the path to that point and where the behavior left expectations.
A basis for evaluation, oversight, and governance, beyond a note that something failed.
Project
Owning backend technical decisions and shaping the AI and agentic approach to purchasing automation.
Backend technical ownership and direction of AI and agentic approaches for procure-to-pay automation at USLBM.
Open the project
Conceptual overview Documents Matching Validation Human review Visibility Documents Matching Validation Human review Visibility
Invoices, purchase orders, and receiving records. They do not arrive in one consistent shape.
Comparing those records. Incomplete or conflicting data is the ordinary case, not the exception.
A result has to be checked before it is treated as reliable.
Some cases need a person. The system should not guess past its uncertainty.
Enough of a record to see what happened when something goes wrong.
Project
Roleplaying where the world, its characters, and your choices continue to matter.
An independent AI-powered roleplaying project focused on a world worth returning to.
Open the project
Conceptual overview Explore Connect Choose Return Explore Connect Choose Return
The aim is a world the player wants to spend time in.
Characters and relationships give the experience personal meaning.
Player choices should feel meaningful within the setting.
The goal is an experience worth continuing across sessions.
Project
From a terrain specification to a checked landscape build.
Tools for creating measured terrain in Cursor and checking Gaea exports for Unreal Engine.
Open the project
Conceptual overview Request Terrain spec Gaea Validation Unreal Not built yet Request Terrain spec Gaea Validation Unreal Not built yet
I describe the landscape in Cursor. That conversation is not the build.
A measured description of the landscape to be created.
Build the landscape and keep the result connected to the reviewed terrain.
Check that the exported assets are consistent before continuing.
The landscape export is prepared for use in Unreal Engine.
Broader asset and scene workflows remain future work.
Project
Diagnostics for systems whose failures are easier to notice than to explain.
A telemetry and observability project for making complex system behavior understandable.
Open the project
Conceptual overview Ingestion Schema Health Dead letters Diagnosis Ingestion Schema Health Dead letters Diagnosis
How telemetry enters the record, and what is lost before it arrives.
Validation and cleanup, so the shape of an event does not drift until it cannot be read.
Whether the signals are arriving, complete, and still trustworthy.
Messages that were not processed, kept as evidence rather than discarded as noise.
A path from a failed signal to an explanation of what the system did.
Project
A place to watch a single-player program, keep the evidence, and separate a guess from an observation.
A local workbench for studying how a single-player program behaves: memory changes, instructions, and the evidence behind a hypothesis.
Open the project
Conceptual overview Observe Evidence Inspect Relate Hypothesis Observe Evidence Inspect Relate Hypothesis
Watch a running single-player process and record what changed. This is the start of a question, not an answer.
Keep observations tied to the session that produced them.
Decode a short window of x86-64 with Capstone. A successful decode is not proof the instruction ran.
Review relationships in collected evidence as questions for further investigation.
Write down what I think a value means, and what would contradict it. There is no verified status.
Project
A gallery that learns what you choose, then answers with a judgement.
A short psychological game on Steam. The art you choose says something about who you are.
Open the project
Conceptual overview Rooms Choices Pattern Judgement Rooms Choices Pattern Judgement
Curated rooms of art. There is no combat and no grind.
What you admire, and the questions that steer how a choice is read.
The gallery pays attention where your preferences contradict one another.
An ending shaped by the decisions you made, not by whether they were the right ones.