Voice AI Trainer
For RZD, we developed an educational platform where employees could study learning topics and take knowledge assessments with an AI assistant via voice interaction — a modern alternative to conventional tests and formal training.
Client
RZD
Period
2 months
Format
Web service with voice assistant
About the project
For RZD, we developed an educational platform where employees could study learning topics and take knowledge assessments with an AI assistant via voice interaction. The project was conceived as a more modern alternative to conventional tests and formal training. Instead of a static scenario with answer choices, the user interacted with the system in a conversational format: studying material, answering questions by voice, and receiving real-time feedback throughout.
The Challenge
We needed to create a digital tool for studying regulatory documentation and conducting knowledge assessments in a more convenient and natural format. For the client, this meant making learning more interactive and assessment more aligned with real communication — where the goal is not simply to select the correct answer, but to demonstrate understanding of the topic and the ability to reason through it.
Our Solution
- Built a web service combining an educational platform with a voice AI assistant
- Implemented speech recognition pipeline using VOSK and Yandex SpeechKit
- Integrated LLM-based response analysis to evaluate answer content and quality
- Built conversational logic: follow-up questions, context retention, and scenario progression
- Developed assessment logic to evaluate not just factual correctness but depth of understanding
Results
- Working voice AI training service delivered in 2 months
- Learning became more interactive — users engage in dialogue rather than selecting from fixed options
- Assessment format moved closer to real communication, testing understanding rather than recall alone
- Regulatory knowledge checks can now be conducted in a flexible, conversational digital format
Similar projects
Анализ сентимента Telegram-каналов
A private client approached us with a hypothesis that Telegram channel publications may influence the behavior of some Moscow Exchange market participants. We built an ML module to analyze messages, detect company mentions, and determine publication sentiment.
Классификация новостей
For AI News, we built an ML microservice for a startup working with a stream of news from multiple sources. The product aggregated news from various channels, analyzed content, and delivered a processed feed without duplicating materials.
Let's buildsomething extraordinary.
Ready to start your next project? Reach out and let's discuss how we can help you achieve your goals.