Tetiana Stepanets grew up in Kyiv, Ukraine, and now works from Nice, France. Before LiliDi AI, Stepanets built Crema Caffe to more than 30 franchise locations and built and exited Homsto, an Amazon DTC brand, in 2024. Stepanets describes herself as a non-technical founder who stays hands-on with testing, UX and implementation.
Lilidi Twin, launched in May 2026, is an AI employee for sales, support and meetings. It talks with customers by video, voice or text, answers from knowledge the business has approved, qualifies leads, books meetings and drafts follow-ups that a person approves before anything is sent. We spoke about why it is a job rather than a chatbot, where it hands over to a human, and why the company is careful not to claim results it has not measured.
Key takeaways
- Stepanets built Twin from her own experience of being on many calls and selling, including selling franchises, and wanted an avatar that works for you in those conversations.
- Twin answers from approved business knowledge, is presented as an AI representative, and hands over to a person when a request is outside its approved role.
- Follow-ups are drafted by the Twin but nothing goes out until a person presses Send.
- LiliDi AI is bootstrapped and early: it currently has 10 pilots and around 120 users, with no paid subscriptions yet.
- The free plan gives 60 live minutes a month by video or voice with no card; over the next 12-18 months the focus is converting pilots into paying customers and making Lilidi Twin reliable in everyday customer conversations.
Before any of this
Startup.talk:Where did you grow up, and what was childhood like?
Stepanets:I grew up in Kyiv, Ukraine.
Startup.talk:What first got you into building things?
Stepanets:Honestly, it came from my own work. I was always on a lot of calls, and I sold a lot, including selling franchises at one point. So I wanted to build an avatar that works for you in those conversations.
“I wanted to build an avatar that works for you in those conversations.”Tetiana Stepanets
Startup.talk:What was your first job, and what did it teach you?
Stepanets:I worked at a fire and rescue station. After college I started working and studied at the same time, getting my higher education in parallel. It taught me to handle all kinds of situations.
The road to Lilidi Twin
Startup.talk:What made you start LiliDi AI, and how did the earlier creative platform lead to Lilidi Twin?
Stepanets:LiliDi AI began with creative AI tools. With Twin, the focus is the customer conversation: helping a business answer a visitor, understand what they need and move to a useful next step. My background is in operating businesses, so I am interested in AI that does a defined job, rather than producing an impressive demo alone.
Startup.talk:You built Crema Caffe and Homsto before LiliDi AI. What did those companies teach you?
Stepanets:They taught me to look at the whole operation. A product, a sale and the service after the sale are connected. Crema Caffe grew to more than 30 franchise locations, and Homsto was an Amazon DTC business that I exited in 2024. At LiliDi AI, that background keeps me close to customer questions, usability and implementation.
Startup.talk:Where do you and the team work from, and how does that shape the way you build?
Stepanets:I am based in Nice, France, and we are a small team. I stay hands-on with testing, UX and implementation. With a small team, we need to keep the work focused on what customers actually need.
Startup.talk:Why an AI employee for sales, support and meetings, rather than a chatbot?
Stepanets:A chatbot describes an interface. We want to describe a job. Twin should answer an enquiry from approved knowledge, ask useful qualifying questions and help arrange the next conversation. Video, voice and text are different ways to do that job.
The product
Startup.talk:Why the name Lilidi Twin?
Stepanets:LiliDi AI is the name of our company. A twin is what the product is: a digital counterpart of you or your business that can talk to customers when you cannot. I wanted a real helper, the way an AI assistant helps you, but with a face and a voice, so the conversation is more engaging.

Startup.talk:Why does Lilidi Twin talk by voice and video as well as text?
Stepanets:People have different preferences and different questions. Sometimes typing is fastest. Sometimes speaking through an explanation is easier. We offer the choice; we do not assume every customer wants an avatar.
Startup.talk:How do you make sure it answers only from what a business has approved, and always says it is an AI?
Stepanets:The business reviews the knowledge and answers its Twin may use, sets its behaviour and defines when it should involve a person. The Twin is presented as an AI representative. Those boundaries need to be tested with real questions before and after launch. We do not treat a natural-sounding answer as proof that it is correct.
Startup.talk:When does Lilidi Twin hand a conversation to a person, and why those cases?
Stepanets:The business defines the handoff rules. A request for a person, a question outside approved knowledge or a decision beyond the Twin's role should not become an invitation to improvise. A useful handoff preserves the context so the customer does not have to start again.
Startup.talk:Why does a person approve follow-ups before they are sent?
Stepanets:An outgoing message represents the business. A person should be able to check what it promises, who receives it and whether the next step is right. Twin can prepare the follow-up, but nothing goes out until the person presses Send.
Startup.talk:What did you learn from letting a Twin join live Zoom, Google Meet and Teams calls?
Stepanets:I would separate the feature from a proven outcome. Twin can join those calls, but we are still early and I do not want to claim quantified results we have not measured. Our focus is making its role clear and keeping the conversation useful, with the business's knowledge and boundaries in place.
Startup.talk:What do you deliberately leave out of Lilidi Twin?
Stepanets:We do not position it as a replacement for every human role. It should not pretend to be a person or make promises beyond the business's instructions. Follow-up approval and human handoff are part of the product, not inconveniences to remove.
Positioning
Startup.talk:Who is Lilidi Twin for, and what job does it do for them? Describe the person who gets the most value.
Stepanets:A founder or small team that receives customer enquiries but cannot be available for every first conversation. Twin provides that first response, explains approved information, qualifies the enquiry and helps book the next step. The aim is to make the handoff to the team more useful.
Startup.talk:How do you view your positioning relative to Tavus, HeyGen interactive avatars, Intercom Fin and Qualified? Is there another competitor you think of first?
Stepanets:I would rather explain our product than make broad claims about competitors. Lilidi Twin combines video, voice and text conversations with approved business knowledge, lead qualification, meeting booking and human-approved follow-up. The question for a customer is which workflow fits their business, not which demo has the most realistic face.
Startup.talk:What is your core moat when bigger platforms keep adding AI avatars and agents?
Stepanets:I do not think an avatar alone is a durable advantage. What we need to earn is trust in the complete workflow: setup, useful answers, clear limits, handoff and follow-up. We are early, so I would describe that as the work ahead rather than claim an established moat.
Startup.talk:What are some of the use cases for Lilidi Twin that are not obvious and can add a lot of value to users? List as many as possible.
Stepanets:Potential uses include explaining a service before a first consultation, answering course enquiries, guiding someone through a product introduction, qualifying a specialist-service request and offering a spoken information point on a kiosk. These are examples of where the workflow could fit, not claims that every use case has already produced measured results.
Startup.talk:What is the one detail in Lilidi Twin you are proudest of?
Stepanets:The human approval step for follow-ups. It is a small interaction, but it makes responsibility clear: AI prepares the work and a person decides what the business actually sends.
The hard part
Startup.talk:How many people use Lilidi Twin today?
Stepanets:We currently have 10 pilots and around 120 users, with no paid subscriptions yet. I would not present the pilots as paying customers or attribute every conversation to a real buyer, because some are tests.
Startup.talk:What is a challenging time you had building Lilidi Twin?
Stepanets:I am a non-technical founder, so a continuing challenge is turning an operational idea into a product that works reliably. I work closely on testing, UX and implementation, and I am learning web coding and AI agents. I do not want to be distant from the details just because I am not the developer.
Startup.talk:Who believed in you before there was anything to show? What did they say or do?
Stepanets:Honestly, nobody did at the start. I believed in it myself, and that was enough to keep building until there was something to show.

What it is all for
Startup.talk:What would make you say Lilidi Twin has succeeded?
Stepanets:When a small business can rely on it for a useful first conversation, and customers feel informed rather than blocked from reaching a person. We also need to prove that people will keep using it and pay for the value it provides.

Startup.talk:You give 60 video minutes a month free with no card. How did you decide what to give away?
Stepanets:The current free plan includes 60 live minutes every month, by video or voice, with no card required. The principle is that someone should be able to test a real conversation with their own business knowledge before deciding whether it is useful enough to pay for more usage.
The road ahead
Startup.talk:What are your milestones for the next 12-18 months?
Stepanets:Over the next 12-18 months, our focus is converting pilots into paying customers and making Lilidi Twin reliable and useful in everyday customer conversations.
Startup.talk:Where do you want Lilidi Twin to be in five years?
Stepanets:I want it to be a trusted first-line colleague for small businesses: easy to set up, clear about its limits and useful across the channels where customers actually arrive. The goal is dependable work, with people still responsible for the business.
Startup.talk:What are your hopes and fears for a post AGI world?
Stepanets:My hope is that AI takes over repetitive work, so small businesses and people spend their time on judgment and real relationships. My fear is people trusting AI blindly, or not knowing when they are talking to one. That is why our avatars always say they are AI and a person stays responsible for what the business says.
“our avatars always say they are AI and a person stays responsible for what the business says”Tetiana Stepanets
This interview was conducted in writing and has been edited lightly for length and clarity.
About LiliDi AI
Lilidi Twin is an AI employee for sales, support and meetings, built by LiliDi AI. It talks with customers by video, voice or text on a website or shared link, in Telegram and in Zoom and Google Meet calls, answers from approved business knowledge, qualifies leads, books meetings and drafts follow-ups for human approval. lilidi.ai
| Attribute | Verified Detail |
|---|---|
| Company | LiliDi AI |
| Founder(s) | Tetiana Stepanets |
| Country | France |
| Grew up | Kyiv, Ukraine |
| Based in | Nice, France |
| Background | Built Crema Caffe to 30+ franchise locations; built and exited Homsto |
| Funding | Bootstrapped |
| Product | Lilidi Twin, launched May 2026 |
More founder interviews
For founders
Building something? Want to share your journey?
Tell us how it started, what nearly broke it and where it is headed. We will turn it into a story like this one.
Share your journeyFor readers
Want to read more human stories behind the startups?
New founder interviews and stories, as they are published.
Follow us on LinkedIn
Lensiq
Decklit
digiQueue
Rewisp