An event-driven AI automation system that receives student homework via Telegram, evaluates it using LLM agents (Ollama / GigaChat), and stores structured feedback and grades in a database with automated notifications.
Manual homework evaluation is time-consuming and does not scale well. Teachers need fast, consistent feedback while students want immediate responses.
This system automates homework evaluation using AI agents triggered by Telegram messages. It integrates messaging, LLM evaluation, structured storage, and notifications into a single automated pipeline.
- Student submits homework via Telegram bot
- n8n triggers a workflow on incoming message
- Student data is created/updated in NocoBase
- Homework text is sent to:
- Local Ollama (offline inference)
- or GigaChat (cloud LLM)
- AI evaluates homework using predefined criteria
- Feedback and grade are:
- sent to the student
- sent to the teacher
- stored in NocoBase
Homework is evaluated using three criteria:
- Understanding
- Application of the lesson tools
- Answer structure
I chose these criteria because they allow to rate answer generally without specifications. As a result, this worflow can be used regardless the theme of assignement.
The AI returns:
- Numeric grade
- Structured teacher-style feedback
- n8n (workflow orchestration)
- Telegram Bot API
- Ollama (local LLM inference)
- GigaChat API
- NocoBase (backend & database)
- JSON / REST APIs
- No API tokens or credentials are stored in the repository
- Personal data is anonymized in examples
- All secrets are managed via environment variables
Actively used in a real educational workflow. Future improvements include:
- rubric-based grading
- plagiarism detection
- analytics dashboard