Tracked messages per session, session length, and whether a session led to a payment.
From a blank page to break-even consumer AI
Gentlyx started as an idea with no defined path. Research found the audience, a pivot moved the bet from people to AI pipelines, and evals set the bar for every release.
Gentlyx / Lovel
- Aug 2024 — Sep 2025
- Fractional CPO
- 12-person team
- Chat AI + image generation
300–400Kmonthly active
users
users
11 moto break-even
~$25Kmonthly revenue
at break-even
at break-even
Starting point
There was an idea and no specific path. Brainstorming inside the team did not produce anything worth building.
There was no product, no team structure, and no evidence yet about who the users would be.
My role
Fractional CPO from the first day. Hired the first people, set up the organization and the delivery process for a 12-person team.
Personally responsible for the chat AI pipelines and the image-generation flow end to end: prompt and model orchestration, output quality, latency and per-request cost.
What I did
- Researched before building
Studied competitors and the audience’s own words on forums, reviews and Trustpilot. The aggregated picture: a core audience of lonely men looking for support and engaging conversation.
- Dropped the two-sided model
The first direction was a classic two-sided product with human hosts entertaining users. It would have been a clone and would not work, so the bet moved to AI pipelines.
- Built a routed chat pipeline
Grok as the conversational model, with a BERT classifier that reads what the user is asking and what kind of user they are, then routes them to the best model.
- Tuned image generation for realism
A Stable Diffusion pipeline, tuned over several iterations to generate realistic people without artifacts or distortions.
- Chose the business model
Subscriptions plus tips.
How we measured it
An image model ships only when 7–8 of 10 generations look real.
For a companion product, a distorted face or an obviously synthetic image breaks the experience instantly, so realism was measured by people before any rollout.
Raters on a crowdsourcing platform flagged artifacts, distortions and any sign that a photo was AI-generated.
Output quality was tracked alongside latency and cost per request for both pipelines.
7–8 / 10realistic generations to release
2pipelines under evals
What changed
Launched 0→1, scaled to 300–400K MAU, and reached break-even after 11 months at approximately $25K monthly revenue. Gentlyx has since been renamed lovel.ai.
300–400KMAU at scale
- M0Launch 0→1
- M11Break-even · ~$25K/mo
Chat AI pipelinesImage generationEvals on both
Summary
Research found the audience before the team built anything, and evals decided what shipped. That combination took a blank page to break-even in 11 months.
“From day one, Mr. Simakov played a central role in shaping our product and growth strategy. Over the course of our collaboration, we went through hundreds of hypotheses across product design, user acquisition, retention and monetization.”