#vacancy #fulltime
**AI/ML Specialist at YaizY
Who we are**
YaizY is an educational platform that helps schoolchildren acquire 21st-century skills and prepare for the digital professions of the future. We create courses in programming, animation, game design, multimedia, and other in-demand areas.
We operate in the US market, offer over 30 courses in 8 career tracks, and collaborate with 100+ schools in 18 states. Over 25,000 students have used our products.
We are now launching a new product from scratch for a new market – AI Math Tutor. This is a personal AI math tutor for schoolchildren that adapts to the student's level, identifies knowledge gaps, and guides them through explanations, practice, and checks to achieve confident mastery of the subject. We are creating a product capable of replacing a traditional math tutor.
What you will do
You will be responsible for everything related to the product's AI logic – from prompts to analytics of learning behavior.
AI Math Tutor is a browser-based web application with a workspace (task + whiteboard + AI chat) where the student solves problems, and the AI tutor analyzes the solution in real-time, diagnoses errors, and guides the student to the correct answer through a multi-level hint system. Details:
- Designing and iteratively debugging prompts for the AI tutor: checking solutions, diagnosing errors, generating pedagogical feedback
- Building an answer validation pipeline: recognizing handwritten solutions from the whiteboard (screenshot → AI → structured assessment)
- Developing a multi-level hint system: from a nudge to a full explanation, in the spirit of the Socratic method
- Classifying student errors: identifying the type (sign, formula, operation, logic, missed step) and generating precise explanations
- Designing routing of requests between models: selecting the optimal model for the task (checking an answer, generating feedback, analyzing a solution) considering cost, speed, and quality
- Building a RAG pipeline for working with educational materials: theory, rules, example problems – the AI tutor needs to retrieve relevant content for a specific topic and student error
- Designing a knowledge graph: mapping dependencies between mathematical concepts (which topics are prerequisites for others, how error types relate to specific skill gaps)
- Configuring and optimizing work with OpenAI API: Assistants / Responses API, Structured Outputs, file search for theoretical materials
Tech Stack
- AI: OpenAI Responses API / Assistants API, Structured Outputs, file search – the primary provider for the MVP stage. We are considering transitioning to alternative models or deploying our own model in the future, including routing requests between multiple models.
- Validation: Zod (server-side AI response validation)
- Math: math.js (deterministic validation of numerical answers)
- Backend: Node.js + Fastify, WebSocket (Socket.io) – for interaction with AI services
- DB: PostgreSQL (attempts, progress, error patterns), Redis (sessions, chat context)
The tech stack may be adjusted. If you have arguments for a different solution, we are open to suggestions.
Requirements
- Full-time, remote work
- Practical experience in prompt engineering for product tasks and iterative prompt debugging with measurable results
- Experience integrating LLMs into products (OpenAI API, Anthropic, or similar): streaming, structured outputs, error handling, fallback strategies
- Understanding of how LLMs work at a level sufficient for making architectural decisions: context window, tokens, temperature, function calling
- Experience working with multiple models and understanding their strengths/weaknesses for task routing between them
- Ability to design data processing pipelines involving AI (input → preprocessing → AI → validation → output)
Will be a decisive advantage
- Experience with OpenAI Assistants / Responses API, file search, Structured Outputs
- Experience building RAG systems or agent scenarios in production
- Experience building knowledge graphs or working with domain ontologies
- Experience deploying and fine-tuning open-source models
- Data analysis skills: ability to build metrics
