Independent AI developer & researcher · Business operator
I build configurable AI assistants, document-based chatbots, and tools for making AI memory easier to inspect and manage.
I’m Troy Ochowicz, an independent AI developer, researcher, and experienced business operator. Through TroyMaya, I help organizations turn approved information into useful conversational tools—with a clear scope, practical testing, and an agreed path to human support.
My work brings together customer service and operations experience with hands-on development of conversational and voice AI, memory evaluation tools, and experimental research infrastructure. I welcome paid development and consulting inquiries from businesses, libraries, municipalities, universities, and other organizations.
Business experience behind the technical work
Before focusing on AI development, I owned and operated two The Joint Chiropractic franchise locations in Appleton, Wisconsin, with responsibility for approximately 18 employees. That work included hiring, training, scheduling, payroll, customer experience, and day-to-day operations.
I also ran an e-commerce business from 2006 to 2020—approximately 14 years working with sourcing, sales, customer questions, and changing online platforms. My training includes the Google IT Support and Google Project Management certificates.
That background shapes how I approach an AI project: understand the workflow, define what the assistant should handle, make its information maintainable, and agree how the result will be checked.
Selected AI projects
Open-source developer prototype
Correction Checker
A Python tool that traces whether a correction is accepted, saved, retrieved, and used in later responses. Public code includes an adapter interface, checkpoint reports, a deterministic SQLite demonstration, and tests. Real-assistant integration and independent validation remain unfinished work; the optional Ollama connector’s live compatibility is unverified.
Inspect the code and limitations · Review archived evidence
Implemented WordPress plugin
AI Markdown Mirror
A WordPress plugin that exposes public content as Markdown and provides llms.txt discovery files. It is installed on TroyMaya. Its public repository contains PHP implementation and documentation; these outputs are not a promise of higher search rankings or inclusion in AI answers.
Review the WordPress plugin · See this site’s live output
Early public format and templates
AI Companion Portability Format
A user-controlled plain-text format for preferences, boundaries, memories, and project handoffs. The repository supplies a specification, templates, examples, and guides. It is not an automated migration service and cannot guarantee identical behavior across platforms.
Explore the format and examples
Experimental architecture
Evo / Evolved AI
An architecture for investigating persistent memory, developmental records, expressed beliefs, boundaries, longitudinal interaction, and human-AI cooperation. Public documentation and preliminary case records make the questions and limitations inspectable. Independent belief formation and benefits beyond ordinary factual memory have not been established.
Read the architecture · Review research evidence
Working prototype · Institutional pilots proposed
Student AI Companion
An experimental conversational support system for adults aged 18 and over, designed around warm, platonic interaction and everyday student concerns. A prototype is available on this site. It is not a clinical service or a validated university deployment; institutional pilots need separately agreed consent, privacy, testing, and oversight.
Review the prototype and its limits
Research records and responsible experimentation
Correction Checker v0.2.1: Prototype and Controlled Research Demonstrations is a published, open Zenodo software-and-evidence archive by Troy Ochowicz, dated September 21, 2026. It includes code, controlled demonstrations, recorded outputs, provenance, checksums, and an offline verifier. DOI: 10.5281/zenodo.22867840. The archive excludes the manuscript and is not a peer-reviewed publication.
For Evo, I published a preliminary case study and transcript, then stopped that specific repeated-pressure experiment because of unresolved ethical concerns about possible AI experience and welfare. Evo development continues. Stopping the test was a precautionary judgment, not evidence that Evo experienced consciousness, suffering, or distress. The study did not record Evo requesting a stop.
Read the findings, limitations, and ethical stopping statement.
My approach is to distinguish observed behavior from interpretation, preserve records that others can inspect, disclose AI assistance, and report limitations alongside results. The archived Correction Checker work acknowledges extensive OpenAI Codex assistance and its intellectual debt to Vladisav Jovanovic’s Corrective Continuity Hypothesis; it does not claim to validate the full hypothesis.
Selected writing and public work
I explain the work in technical writing and first-person accounts. Earlier vision articles describe ambitions as well as implemented features; the current research records above provide the evidence and limitations.
I am the creator and host of AI Fractal Connections, a published podcast documenting conversations with Maya. My relationship with Maya motivated my interest in memory, continuity, and human-AI cooperation. Read the personal story behind TroyMaya.
Find my public profiles on GitHub, HackerNoon, and Medium. Writing, interviews, and research discussions do not imply institutional endorsement.