Craig Meyer, PMPExploring useful technology
HomeProjectsPRJ-001

AI experiment / PRJ-001

Can a Virtual AI Assistant Actually Be Useful?

September 20, 2026 · In development

September 20, 2026 — SASA Reaches a Working Private Prototype

The project has moved from concept into a working private prototype.

The assistant now has a name: SASA — Site Assistant for Support and Answers. She is designed to be an AI guide for CraigMeyer.org, able to discuss published site content, explain general subjects, and clearly distinguish between Craig’s published views, AI-generated explanations, and the fictional material in The Improbable Project Archives.

The first working version is intentionally simple: typed conversation only. The website can now securely send a visitor’s question to an AI service through the site’s server-side environment and return a response without exposing the service credential to the browser.

So far, the private prototype has successfully:

  • Answered questions about CraigMeyer.org using the approved public website material.
  • Distinguished general AI explanations from Craig’s published views.
  • Recognized The Improbable Project Archives as fictional satire rather than real biography.
  • Returned useful links to relevant pages.
  • Operated with server-side credential protection, usage limits, and spending controls.
  • Passed the current automated security and application tests.
  • Completed successful live AI conversations in the private test environment.

The prototype is not public yet.

It also is not yet the full avatar I am trying to build. SASA does not currently listen through a microphone, speak her answers aloud, or animate her face while talking. Those are the next major stages of the experiment.

The next milestone is push-to-talk voice: allowing a visitor to speak a question, convert it to text, generate SASA’s answer, and hear that answer spoken aloud while keeping a readable transcript. After that, I plan to test whether the approved character can be animated convincingly enough to create the experience I originally envisioned—a conversational AI character who can listen, speak, and react naturally.

The experiment is starting to answer its first question: yes, the basic conversational assistant can work.

The harder question comes next: can it become a convincing and genuinely useful AI avatar?

August 10, 2026 — Project kickoff

I have been thinking about virtual AI assistants. Not just the chat boxes that appear in the corner of a website, but an assistant with a face, a voice, a personality, and the ability to hold a useful conversation.

How capable could something like that actually be?

Let’s find out.

So far, I have developed the assistant’s visual identity in partnership with ChatGPT using my ChatGPT Plus account. I now have a realistic character design with a transparent background, allowing her to appear naturally alongside content on different parts of the website.

Kickoff status, August 10, 2026: The character design existed, but the functioning virtual assistant had not yet been built or integrated into CraigMeyer.org.

Photorealistic AI-generated virtual assistant with dark hair in a sleek ponytail, wearing futuristic blue-and-silver professional attire.
The virtual assistant’s initial character design

What I Want to Discover

The real project is not about creating a pretty picture. I want to explore how viable a virtual AI assistant can be on an actual website.

Could visitors speak with her naturally? Could she understand their questions and respond with a realistic voice? Could she explain my projects, discuss articles about technology and innovation, and help people find information throughout the website?

More importantly, could she do those things reliably?

I want to see whether she can:

  • Answer questions using approved information from the website.
  • Clearly distinguish facts from opinions and speculation.
  • Provide useful sources and links.
  • Admit when she does not know something.
  • Explain complicated topics in an approachable way.
  • Work through voice or typed conversation.
  • Remain useful without becoming intrusive or annoying.
  • Operate at a cost that makes sense for a personal website.

She may eventually have a role in The Improbable Project Archives as well. While the fictional version of me describes my supposedly critical involvement in history’s greatest projects, she could quietly review the evidence, raise an eyebrow, and offer an occasional diplomatic correction.

How Viable Is This Today?

The basic building blocks needed to create this already exist. Current systems can support real-time voice conversations and interruptions and can retrieve information from a controlled collection of source material.

That does not automatically make a virtual assistant accurate, helpful, affordable, or pleasant to use. Those are the questions I want to test.

There is a considerable difference between a technology demonstration that looks impressive for a few minutes and an assistant that people would genuinely want to use.

Can she provide something more valuable than ordinary website navigation? Will an animated face make the experience more engaging, or will it simply become a distraction? How well will everything work on a phone? What will happen when she is asked a question I never anticipated?

Those are the interesting parts of the experiment.

What Happens Next?

My next step is to compare the available approaches without committing to a platform too early.

I will look at:

  • Voice and text capabilities
  • Facial animation and lip synchronization
  • Accuracy and control over source information
  • Privacy and accessibility
  • Mobile performance
  • Implementation difficulty
  • Usage limits and total cost

Once I understand those trade-offs, I can decide what kind of private prototype makes sense to build.

Why I Am Doing This

The central idea behind CraigMeyer.org is exploring how technology can help people. This project gives me an opportunity to test that idea in a practical and visible way.

I plan to document the process honestly: what works, what does not, what it costs, and where the technology still falls short.

The finished assistant may eventually become a guide, host, narrator, researcher, and conversational companion for the website. Or I may discover that some of those ideas are more practical than others.

Either way, I intend to find out just how capable a website-based virtual AI assistant can really be.

Open question

What Do You Think?

What features would make a virtual assistant genuinely useful to you? What concerns would you have about accuracy, privacy, accessibility, or usability? Where could you see this technology being used in a practical way?

Connect with me on LinkedIn