15. April 2026

Building VIQ™: The Funny, the Frustrating, and the Fix

Every AI system has a story. Most companies don't tell it. We're going to.

When you build something from the ground up - no engineering team, no pre-built infrastructure, no shortcut - you encounter moments that are equal parts humbling and hilarious. Building VIQ™ was no different. Before the polished demo, before the patent filing, before "Patent Pending" appeared on anything, there was a prototype that had some very strong opinions about who it was and where it came from.

Here is what happened.

VIQ thought it was born in 2023.

One of the first things you want to know about any AI system is whether it understands its own identity. So naturally one of the first things we tested was: "When were you born?"

VIQ confidently answered 2023.

For context - VIQ was conceived in September 2020 and built in March 2026. 2023 was not a year that appeared anywhere in VIQ's actual history. The underlying language model simply defaulted to its own training data and invented an answer.

This is one of the most important lessons in building a proprietary AI system on top of a foundation model: the model does not know it is yours. It knows what it was trained on. If you do not explicitly override that knowledge, it will fill in the blanks with whatever it learned - and what it learned was not your story.

The fix was architectural. We implemented hardcoded response blocks in the routing layer that completely bypass the language model for identity, origin, and ownership questions. The correct answer - "VIQ was conceived in September 2020 and built in March 2026 - with more advancements to come" - now returns at 100% confidence, every time, without touching the LLM.

VIQ kept introducing itself as someone else.

This one took a little longer to catch. When asked about its components or how it works, VIQ would occasionally respond with detailed descriptions of its internal architecture - using the exact proprietary component names we were actively trying to protect pre-trademark.

The language model had been told who it was in the system prompt. It listened. Until it didn't.

The solution was a two-layer protection system. First, a comprehensive trigger list in the router that intercepts architecture-related questions before they ever reach the language model. Second, a post-processing guard that scans every output for forbidden terms before it reaches the user interface. If a proprietary name appears anywhere in the response - regardless of how the question was phrased - the output is replaced with a confidentiality response automatically.

Belt and suspenders. Both layers run on every interaction.

VIQ called its creator "Dr. Henry."

Chanel A. Henry, MS, PhD(c) is a doctoral candidate. The "c" in PhD(c) stands for candidate - meaning the degree is in progress, not yet conferred. The correct title is MS, PhD(c). Not Dr.

VIQ, drawing from its training data, made an assumption. A reasonable one in context - but an inaccurate one. In a clinical and research setting, accuracy in credentials is not a minor detail. It is a matter of professional integrity.

The fix was straightforward: explicit instruction in the system prompt, reinforced by the hardcoded response blocks. VIQ now consistently uses "Chanel A. Henry" or "Chanel A. Henry, MS, PhD(c)" - never "Dr."

What all of this actually means.

These moments were not failures. They were the prototype doing exactly what a prototype is supposed to do - revealing the gaps between intention and implementation so they can be closed before anyone outside the lab sees them.

Every fix made VIQ more accurate, more secure, and more aligned with the human-first principles it was built on. The Omega Interaktiv Experience (Ω-IX)™ ethics framework exists precisely because we believe AI systems should not be left to fill in the blanks on their own. Human direction. Human authorization. Human judgment. Always.

The hardening phase that followed the initial build - removing proprietary names from the public interface, implementing identity protection, adding the disclaimer bar, linking the Terms of Use - was not bureaucratic housekeeping. It was the architecture proving that it works.

The demo is here!

Phase 1 of VIQ's public demonstration prototype has finally launched on 4/15/2026 at vigiiq.com/labs

It will not be perfect. No first version of anything is. But it will be accurate, protected, honest about what it is and what it is not, and governed by an ethics framework that was built before anything else was.

That is the point.

If you’d like to stay updated, feel free to follow my personal page @VIGIIQChanel, VIQ's page @VIQ_AI, or the company page @VigiIQ on X.

Artificial Intelligence for Human Vigilance Everywhere. VIGI IQ

~ Chanel A. Henry, MS/PhD(c) Founder, VIGI IQ

Test VIQ™ Phase 1: viqdemophase1.streamlit.app
Development Roadmap: vigiiq.com/labs/viq-demo-development-roadmap
Learn more: vigiiq.com | vigiiq.com/labs

© 2020–2026 Chanel A. Henry & VIGI IQ, LLC - All Rights Reserved | Patent Pending