AI Story Chat: an engagement experiment for a 4M+ MAU platform by Irvin YakovenkoAI Story Chat: an engagement experiment for a 4M+ MAU platform by Irvin Yakovenko

AI Story Chat: an engagement experiment for a 4M+ MAU platform

Portfolio project
Irvin Yakovenko
AI Story Chat: an engagement experiment for a 4M+ MAU reading platform
Problem. A market-leading reading platform (4M+ MAU) wanted to grow Time Spent. The question was whether modern LLM tools could create a new way into the stories.
Role. Product lead and design engineer: hypothesis, LLM architecture, business case, A/B design, prototype.
Process.
Framed the concept: a side product where readers replay a novel's plot as themselves. They talk to characters and move through chapters.
Designed the experience. Progress is saved and readers can switch characters. They pick ready-made actions or type their own lines.
Built the prototype in React and Next.js with Tailwind, shadcn and a custom design system.
Wrote a book parser and a Go/TypeScript data pipeline. It extracts story data and generates text, images and database seeds.
LLM layer on Genkit with OpenAPI. I wrote prompts that keep the storyline and handle character lines.
Designed a full A/B experiment. The main metric was Time Spent per user. I calculated the MDE from current revenue and Time Spent, targeting 2.5M RUB of extra profit per week.
Outcome. Users who adopted the chat spent 100 minutes on average, against 131 minutes across the platform. Adoption was 10%, too low to move platform-wide engagement. The team kept the feature and changed how it is promoted. The test stopped a blind rollout and pointed to distribution as the next problem to solve.

0 likes · 0 views · Nov 5, 2026

An LLM story chat for a leading reading platform, validated with an A/B test before rollout.

Tags