Projects

A selection of things I've built—across ML research, behavioral simulation, and product. For experiments that didn't work out, see Trials & Errors.

Currently building · Private pilot

Archeode

A product-intelligence platform with one promise: never lose the “why” behind your product. It learns from the work teams already do—tickets, PRs, Slack, docs—to keep a living map of decisions and surface impact analysis before changes ship. No blank-page wiki upkeep.

archeode.com →

Research & ML

A Holographic Brain for LLMs

A fixed-size associative memory for language models, based on Holographic Reduced Representations. Hit 100% recall at 2,600 facts with a ~42× compression ratio versus a naive vector database—after debugging a catastrophic cross-talk failure down to embedding orthogonality.

HRR · NumPy · FAISS · SentenceTransformers

Persona clusters from the behavior simulation

A Sandbox for Human Behavior

300 AI personas with distinct personalities and ideologies, exposed to 3,000 real tweets, producing 3,317 engagement events with realistic power-law dynamics and traceable decision paths—a research instrument for asking why people engage (or stay silent).

Agent-based simulation · Stochastic cognition · Networks

The Valet-Park reinforcement learning environment

Valet-Park: RL Under Pressure

Teaching a reinforcement-learning agent the multitasking puzzle of valet parking—spatial reasoning, memory, and time pressure—to see whether it discovers human-like heuristics or invents its own.

Reinforcement learning · Pygame / Unity

Object Recognition With Help From Background

Peer-reviewed work (ICLR 2022, CSS Workshop) on when and how image backgrounds help object classification—separating class-dependent from class-independent factors to explain non-trivial background-only accuracy.

Explainable AI · ICLR 2022 Workshop

Product

Privacy-Centric Survey Analytics

Algorithms that keep individual respondents anonymous—k-anonymity, l-diversity, t-closeness, and differential privacy—without gutting the analytical usefulness of the data, using dynamic generalization hierarchies tuned to the data distribution.

SHAPE · Privacy engineering

SHAPE proprietary scoring system

Proprietary Scoring System

A scoring methodology that moves beyond simple averages to a more nuanced, holistic view of employee feedback across 12 measurement dimensions.

SHAPE · Measurement design

Flourishing Measurement Framework

A framework—built as a founding member of the World Flourishing Organization—to identify and measure flourishing inside organizations, combining internal data with external signals into a single standard.

World Flourishing Organization · Framework design