People often talk about AI’s potential, but rarely do they share real, measurable results. I want to break that pattern by showing you stats, behind-the-scenes insights, and student reactions to the integration of AI into an educational platform. This will be useful if you’re looking to cut teacher-related costs, improve your SLA, and boost student retention.
Hi there! In Jan of 2023, we launched EvaHelp — or “Eva, help me!” — an AI-powered assistant for educational projects. Since then, Eva has been integrated into the learning processes at karpov.courses, ProductStar, Kodland, Careerum, Kampus, and many others.
We’ve been working with karpov.courses since the very start, and we finally managed to collect measurable data — and we can’t wait to share the results.
What is EvaHelp
Eva is a ChatGPT-based AI assistant. We created her to optimize teacher workloads in online schools. Before Eva, my co-founder and I worked in other EdTech companies and saw firsthand how many questions students sent to instructors — and how resource-intensive it was to respond to them all.
Our hypothesis: ChatGPT could handle a large portion of those questions. Online schools could cut back on resources while students still receive high-quality support — available 24/7. Of course, education is nuanced, but we aimed to cover the most basic case: “I don’t know how to solve this assignment — help!”
How karpov.courses Handled Questions Before EvaHelp
The school offers both free and paid courses. Over 3,000 students study on paid courses each month, and all of them could reach out to instructors for support. Instructors answered every query manually.
Students in free courses also got answers — though there were fewer requests. They could reach out via email or Discord.
Back then, no one tracked the volume of queries. We only started gathering this data a year later to compare it with post-Eva results.
How EvaHelp Was Integrated at karpov.courses
karpov.courses was among the first to implement our product — they got the earliest version of Eva: a barebones ChatGPT widget. There was no dashboard, no response ratings — just a piece of code they had to embed in their platform, and a very roughly designed widget.
They rolled it out to a small group of students. Feedback showed that Eva genuinely helped them learn. One by one, course managers asked to implement Eva in their programs.
What’s great is that Eva can be customized per course — you can set different system prompts or pre-widget student hints depending on the content.
At the same time, we were already thinking about letting schools “train” Eva with their own knowledge. That feature came later, with support for uploading custom files (almost unlimited in size). But funnily enough, karpov.courses didn’t even need that — the standard ChatGPT model performed well on programming and analytics questions.
Eva improved thanks to student feedback, which they shared consistently. We could see what they liked — and what needed improvement — and adapted the product to match user needs.
How Students Reacted to Eva
Most students were happy to get help from Eva and rated her positively.
They could rate responses right in the widget and leave detailed feedback explaining what Eva did well — and what she didn’t.
Course managers could see the percentage of downvotes and even specific comments on what users liked or disliked.
Course managers could see the percentage of downvotes and even specific comments on what users liked or disliked.
I asked why Eva caught on so well at karpov.courses. You’d think a chatbot would feel less valuable than a human expert — but that wasn’t the case.
"Here’s the key: we didn’t reduce human support at all. Students could still ask just as many questions. What we did do was emphasize that Eva, unlike humans, is available 24/7. And — working with GPT is now a critical skill for many jobs, so students were excited to build that experience as part of their learning."
— Ksenia Barkovskaya, Student Support Lead at karpov.courses
Eva isn’t hidden — she’s integrated into the course programs and even has her own section on the site.
Challenges We Faced
OpenAI’s API was blocked for Russian users at one point. We quickly set up stable proxy servers to ensure Eva stayed connected to ChatGPT updates. This workaround is reliable and helps us stay operational despite restrictions.
Another issue: on mobile and tablets, the widget used to cover parts of the content and couldn’t be minimized. Students called this out, and we fixed it — now Eva can be hidden when needed.
Overall, things went smoothly. And ensuring Eva’s stability remains a top priority for us.
What We Achieved in One Year
Over 30,000 students used Eva across both free and paid courses.
Eva directly improved engagement — students stayed on the school’s platform for help instead of turning to outside services.
Instructors now use Eva too — to generate content. Since Eva gives them direct access to ChatGPT (without needing third-party tools), they’ve optimized their workflows as well.
Student interest in interacting with AI has grown. We’ve seen a sharp increase in usage, and it shows in the numbers:
Conversion to a second session (in weekly cohorts of new users) rose from 31% in summer 2023 to 52% in summer 2024.
The time between first and second sessions dropped from 1.72 days (Sept 2023) to 0.89 days now. Students are coming back twice as fast.
All these engagement metrics can be tracked in the Eva admin dashboard.
Most importantly: the number of questions handled by human instructors dropped by 21.35%. Translate that into business terms, and the savings speak for themselves.
"To calculate that number, we gathered data on the average number of students, questions asked, and expert availability before Eva’s launch. We modeled the correlation between those factors and forecasted what the numbers should’ve been based on student growth — then compared that to actual post-Eva numbers. Student count grew by 172.97%. Question volume grew by 115.76%. That delta let us reduce expert workload by 21.35% — a massive optimization. Sure, it’s not a pure forecast — it doesn’t factor in seasonality — but it reflects the real situation quite well."
— Ksenia Barkovskaya, Student Support Lead at karpov.courses
What Else Eva Can Do
Over the past year, Eva has evolved beyond just being a plain GPT model:
- Teacher Mode: Eva doesn’t give direct answers, but instead offers guiding questions and hints.
- You can upload your own knowledge base.
- Eva can be seamlessly integrated and branded to match your platform’s UI.
- Coming soon: Eva will be able to run student surveys — perfect for quick tests and feedback collection.
Eva grows with each new client — we actively respond to requests and ship new features regularly.
How to Connect EvaHelp
Connecting Eva is as easy as adding a Facebook Pixel:
- Register at EvaHelp.ai
- Create a project with your domain
- Embed the code into your site
- Voilà! You’re ready to test Eva.
Best part? You don’t need heavy dev resources — even a project manager can do the integration.
Still have questions? I’d be happy to answer them on a demo — just grab a slot in my calendar.
It’s not the most structured thing, but we do share updates and thoughts on AI in education in our Telegram channel: “AI in Education — Help Me, Eva!”
Hi there! In Jan of 2023, we launched EvaHelp — or “Eva, help me!” — an AI-powered assistant for educational projects. Since then, Eva has been integrated into the learning processes at karpov.courses, ProductStar, Kodland, Careerum, Kampus, and many others.
We’ve been working with karpov.courses since the very start, and we finally managed to collect measurable data — and we can’t wait to share the results.
What is EvaHelp
Eva is a ChatGPT-based AI assistant. We created her to optimize teacher workloads in online schools. Before Eva, my co-founder and I worked in other EdTech companies and saw firsthand how many questions students sent to instructors — and how resource-intensive it was to respond to them all.
Our hypothesis: ChatGPT could handle a large portion of those questions. Online schools could cut back on resources while students still receive high-quality support — available 24/7. Of course, education is nuanced, but we aimed to cover the most basic case: “I don’t know how to solve this assignment — help!”
How karpov.courses Handled Questions Before EvaHelp
The school offers both free and paid courses. Over 3,000 students study on paid courses each month, and all of them could reach out to instructors for support. Instructors answered every query manually.
Students in free courses also got answers — though there were fewer requests. They could reach out via email or Discord.
Back then, no one tracked the volume of queries. We only started gathering this data a year later to compare it with post-Eva results.
How EvaHelp Was Integrated at karpov.courses
karpov.courses was among the first to implement our product — they got the earliest version of Eva: a barebones ChatGPT widget. There was no dashboard, no response ratings — just a piece of code they had to embed in their platform, and a very roughly designed widget.
They rolled it out to a small group of students. Feedback showed that Eva genuinely helped them learn. One by one, course managers asked to implement Eva in their programs.
What’s great is that Eva can be customized per course — you can set different system prompts or pre-widget student hints depending on the content.
At the same time, we were already thinking about letting schools “train” Eva with their own knowledge. That feature came later, with support for uploading custom files (almost unlimited in size). But funnily enough, karpov.courses didn’t even need that — the standard ChatGPT model performed well on programming and analytics questions.
Eva improved thanks to student feedback, which they shared consistently. We could see what they liked — and what needed improvement — and adapted the product to match user needs.
How Students Reacted to Eva
Most students were happy to get help from Eva and rated her positively.
They could rate responses right in the widget and leave detailed feedback explaining what Eva did well — and what she didn’t.
Course managers could see the percentage of downvotes and even specific comments on what users liked or disliked.
Course managers could see the percentage of downvotes and even specific comments on what users liked or disliked.
I asked why Eva caught on so well at karpov.courses. You’d think a chatbot would feel less valuable than a human expert — but that wasn’t the case.
"Here’s the key: we didn’t reduce human support at all. Students could still ask just as many questions. What we did do was emphasize that Eva, unlike humans, is available 24/7. And — working with GPT is now a critical skill for many jobs, so students were excited to build that experience as part of their learning."
— Ksenia Barkovskaya, Student Support Lead at karpov.courses
Eva isn’t hidden — she’s integrated into the course programs and even has her own section on the site.
Challenges We Faced
OpenAI’s API was blocked for Russian users at one point. We quickly set up stable proxy servers to ensure Eva stayed connected to ChatGPT updates. This workaround is reliable and helps us stay operational despite restrictions.
Another issue: on mobile and tablets, the widget used to cover parts of the content and couldn’t be minimized. Students called this out, and we fixed it — now Eva can be hidden when needed.
Overall, things went smoothly. And ensuring Eva’s stability remains a top priority for us.
What We Achieved in One Year
Over 30,000 students used Eva across both free and paid courses.
Eva directly improved engagement — students stayed on the school’s platform for help instead of turning to outside services.
Instructors now use Eva too — to generate content. Since Eva gives them direct access to ChatGPT (without needing third-party tools), they’ve optimized their workflows as well.
Student interest in interacting with AI has grown. We’ve seen a sharp increase in usage, and it shows in the numbers:
Conversion to a second session (in weekly cohorts of new users) rose from 31% in summer 2023 to 52% in summer 2024.
The time between first and second sessions dropped from 1.72 days (Sept 2023) to 0.89 days now. Students are coming back twice as fast.
All these engagement metrics can be tracked in the Eva admin dashboard.
Most importantly: the number of questions handled by human instructors dropped by 21.35%. Translate that into business terms, and the savings speak for themselves.
"To calculate that number, we gathered data on the average number of students, questions asked, and expert availability before Eva’s launch. We modeled the correlation between those factors and forecasted what the numbers should’ve been based on student growth — then compared that to actual post-Eva numbers. Student count grew by 172.97%. Question volume grew by 115.76%. That delta let us reduce expert workload by 21.35% — a massive optimization. Sure, it’s not a pure forecast — it doesn’t factor in seasonality — but it reflects the real situation quite well."
— Ksenia Barkovskaya, Student Support Lead at karpov.courses
What Else Eva Can Do
Over the past year, Eva has evolved beyond just being a plain GPT model:
- Teacher Mode: Eva doesn’t give direct answers, but instead offers guiding questions and hints.
- You can upload your own knowledge base.
- Eva can be seamlessly integrated and branded to match your platform’s UI.
- Coming soon: Eva will be able to run student surveys — perfect for quick tests and feedback collection.
Eva grows with each new client — we actively respond to requests and ship new features regularly.
How to Connect EvaHelp
Connecting Eva is as easy as adding a Facebook Pixel:
- Register at EvaHelp.ai
- Create a project with your domain
- Embed the code into your site
- Voilà! You’re ready to test Eva.
Best part? You don’t need heavy dev resources — even a project manager can do the integration.
Still have questions? I’d be happy to answer them on a demo — just grab a slot in my calendar.
It’s not the most structured thing, but we do share updates and thoughts on AI in education in our Telegram channel: “AI in Education — Help Me, Eva!”