Harnessing the full potential of Conversational AI to revolutionize communication, training and assessment.
TalkLab builds Parallels: human-like voice agents with emotional and cognitive awareness that teach, simulate and assess conversations the way an expert would.
“Conversational science is the behavioral and cognitive study of how humans talk, listen, and coordinate meaning.”
Partners
The TalkLab Studio
The goal is to create the most human-like AI agent possible, what we call Parallels.
- Parallels
- They are capable of having lifelike conversations based on a prompt or an uploaded knowledge base.
- Events
- We can then place host-style Parallels into structured real-world conversations, called Events, that produce reliable data usable to administer an education-based oral assessment, a survey, training, job interview or any scenario you choose. Events assess and gather data on participating users, creating the most robust analytics you can get from any conversation.


We are Educators building for Education, but educational tools can be used everywhere. What we call K‑to‑Corporate.
A lot of our team's reputation is in learning and assessment, so we start there. From kindergarten to corporation, our Parallels and Events System give people a conversational experience, and within the Studio we can derive specific analytics on the fly and as detailed reports.
- K–12
- College & university
- Graduate
- Professional
- Corporate
Questions we hear
What is TalkLab?
TalkLab is a leader in human-like cognitive conversational AI. We tune models, develop proprietary emotion systems and engineer the application layers that make human-like AI agents, called Parallels, and we make TalkLab Studio for building them and running them as Events.
What is a Parallel?
A Parallel is a TalkLab AI agent with its own knowledge, features and personality, plus our patent-pending emotion and rapport emulation system. It can act as a companion or as an interrogator: a tutor, an interviewer, a patient, a customer or an examiner.
What is an Event?
An Event is a structured, real-world conversation built in TalkLab Studio: an oral assessment, a survey, an interview, a pitch or a debate. People take part through TalkList, and the Studio produces analytics on the fly and as detailed reports.
How is a Parallel different from ChatGPT or a chatbot?
TalkLab's application layer strips the agreeable, sycophantic behaviour out of standard language models. Its agents may disagree with you, and say so. Parallels also track the signals experts watch for (uncertainty, confidence, tension, readiness) and give diagnostic feedback.
Can it replace multiple-choice tests?
That is what our first pilots are for. Oral-assessment Parallels are being piloted with graduate institutions as an alternative to multiple-choice questions. Talking with a Parallel shows how well someone understands the material, not only whether they can recognize the answer.
How long does it take to set up?
About 5 minutes to deploy a live Parallel (TalkLab average user data). A live voice agent typically takes 2–12 weeks (Retell.ai, 2025).
Does it work by voice?
Yes, and by text. Parallels speak turn-based today, with full-duplex conversation coming soon.
Who is behind TalkLab?
TalkLab's CEO is Dr. Steve Joordens and its Chief Research Officer is Dr. Brian Harrington, both professors at the University of Toronto Scarborough, with CTO Robin Walker and COO Sean Ramsay. TalkLab's engineers and scientists are building a functional cognitive architecture for Parallels within the Vector Institute's DaRMoD accelerator.
Join us.
Book a demo, bring a pilot, or talk to us about our research.
