X5 Tech shared how it built an AI service for international exam preparation in 7 days

The ExamLab Bot team at the AI Talent Camp intensive assembled a working AI service in a week for international exam instructors. The system transforms syllabuses and deadlines into a personalized learning trajectory and reduces preparation planning time from 3–4 hours to approximately two minutes.

What Was the Problem

The idea grew from a very specific pain point for IB and A-Level instructors. To take on a new student, they need more than just reading the syllabus—they need to turn it into a detailed plan spanning 20–30 weeks, accounting for the exam deadline, schedule, topics already covered, and current level. If a lesson gets rescheduled or a student falls behind, the plan has to be recalculated from scratch.

About four hours of routine preparation work goes into each student, and it's this administrative work, not the teaching itself, that limits growth. As a result, many instructors don't create full personalized trajectories at all: some teach by intuition, some use a one-size-fits-all template, some simply refuse new students. For a segment where tutoring from a top expert costs 30,000–40,000 rubles per month, this represents a direct loss of revenue.

The ExamLab Bot team decided to automate precisely the repetitive organizational layer: plan construction, constraint management, progress updates, and goal tracking.

How They Built the MVP

Over seven days, the team deliberately narrowed the scope to two key functions: building a personalized trajectory for the student's goal, deadline, and level, then adjusting it as the student progresses and the instructor provides feedback. They chose a Telegram bot as the interface to avoid spending time on heavy web frontend development. Google Sheets became the working representation of the plan, and scheduling was synchronized through Google Calendar. As a result, instead of 3–4 hours of manual preparation, plan generation takes about two minutes.

• instructor or administrator sets the goal, deadline, level, available time, and priorities
• the bot generates a detailed learning plan for the entire period
• the plan is saved to Google Sheets, and events are added to Google Calendar
• as progress changes, AI recalculates the trajectory without rebuilding from scratch

Technically, the service was built around a multi-agent scheme. A Python orchestrator coordinates agents with different roles: one handles strategy and time allocation, another handles lesson details, homework, and tests, a third handles batch processing of long programs, and a separate layer validates structure, dates, and coverage completeness. The stack used Python 3.11+, asyncio, aiogram, SQLAlchemy, PostgreSQL, and the OpenRouter API. They abandoned LangChain and LangGraph in favor of direct API calls: this makes it simpler to control prompts, retry logic, and development speed.

"A working minimal product with a clean pipeline is more valuable than

trying to do everything."

Why They Finished in a Week

The intensive itself was structured as a short product cycle, not a typical hackathon with demos for the sake of demos. Day by day, the team went from discovery and problem statement through risk assessment and PoC to MVP, user feedback, and final presentation. Mentors constantly reminded them that the goal was not a polished stage demo but a foundation for a real AI product that could be developed further. This forced quick architectural decisions, elimination of unnecessary features, and a focus on working scenarios rather than flashy ideas.

The project authors specifically emphasize that speed didn't come from AI tools working magic on their own. The interface, interaction scenarios with instructors, and even the choice of initial user segment changed several times during the week. Betting on a minimal working version and rapid testing on real scenarios turned out to be more useful than trying to design the perfect system in advance. The next step is closed beta testing with 10–15 instructors, followed by expanding the exam offerings and adding Russian high school exit exam (ЕГЭ) preparation.

What This Means

The ExamLab Bot case shows that in EdTech right now, narrow AI products with clear value metrics work best. Here, that metric was time: hours of manual preparation became minutes, and instructors gained the opportunity to scale their practice without increased administrative burden. This is an important lesson for other teams too: simple interfaces, direct integrations, and frequent user validation often deliver more than a complex tech stack and a long feature list.