I build reliable AI software systems.

I build systems that automate decisions, manage state, and turn complex workflows into reliable products.

Selected Work

Featured Projects

Flagship systems shown through the problem, challenge, stack, and evidence you can inspect.

How does a run become trustworthy territory state?

PROJECT 01

Product Engineering

A location-based territory strategy game where running changes a shared world.

Tech: Flutter / Python / SQLite / GPS
Problem: Make outdoor runs change a persistent shared strategy world.
Challenge: Spatial systems, anti-cheat logic, synchronization, and long-term world simulation
Evidence: Territory map, capture flow, anti-cheat signals, and sync model

Can the reasoning path be audited after the decision?

PROJECT 02

Decision Learning

A research platform that learns from decision quality over time through auditable reasoning, behavioral data, and continuous evaluation.

Tech: Python / LLMs / Behavioral Modeling / Decision Analysis
Problem: Improve decision quality without hiding evidence behind AI output.
Challenge: Auditable reasoning, behavioral data, continuous evaluation, and evidence preservation
Evidence: Reasoning graph, evidence chain, attribution model, and evaluation loop

Which strategy generation fits which market regime?

PROJECT 03

Trading Systems

A rule-based trading agent evolved into a modular decision architecture focused on explainability, evaluation, and continuous improvement.

Tech: Python / Pandas / Backtesting / Technical Analysis
Problem: Move from trading signals to reviewable decision pipelines.
Challenge: Explainability, evaluation, continuous improvement, and modular decision architecture
Evidence: Strategy timeline, backtest curve, and decision pipeline

Can generated media remain editable after rendering?

PROJECT 04

AI Infrastructure

A local-first AI video clipping and captioning platform for turning long-form video into editable short-form outputs.

Tech: Python / SQLite / Groq Whisper / OpenCV
Problem: Keep AI-generated clips editable after transcription and rendering.
Challenge: Video ingestion, transcription, clip selection, smart crop, captions, and editable render state
Evidence: Caption timeline, transcript state, AI workflow, and render recovery

Process

I build systems that keep evidence close to the decisions they make, so the work can be questioned, improved, and trusted under changing conditions.

My default loop is constraint, hypothesis, instrument, evaluate, then revise.

Constraint: Define the real boundary: data quality, latency, user trust, or failure mode.
Instrument: Add traces, tests, metrics, or review points before treating a system as solved.
Revise: Use evidence to change the architecture instead of defending the first version.