Client's task:
A personal tool was needed for learning German words, which operates on the principle of spaced repetition, where words are repeated not consecutively, but depending on how well you know them.
Solution:
I developed a Telegram bot based on n8n and Supabase. The user sends a German word - the bot automatically translates it via GPT and saves it with the status "New." Then, a system with three testing modes operates with a smart change of statuses: a correct answer advances the word, while an incorrect one returns it for review. After each answer, regardless of the result, GPT generates 3 example sentences with this word and its breakdown - the word is reinforced through context, not rote memorization. New → Semi-new → Learned. Words with the status "Learned" are no longer shown, and mistakes are highlighted in a separate mode and processed separately.
Result for the business:
The user focuses only on those words that truly need attention, does not waste time on what they already know, and does not miss those moments where mistakes are made. The system automatically adapts to each user. The bot is easily customizable for any language - just change the prompt.
Numbers:
- 3 testing modes: new words, mistakes, repetitions
- 4 word statuses: New → Semi-new → Learned / Mistake
- Customizable for any language without changing the architecture
Stack:
n8n · Telegram Bot API · GPT-4o Mini · Supabase · JavaScript Code Nodes
A personal tool was needed for learning German words, which operates on the principle of spaced repetition, where words are repeated not consecutively, but depending on how well you know them.
Solution:
I developed a Telegram bot based on n8n and Supabase. The user sends a German word - the bot automatically translates it via GPT and saves it with the status "New." Then, a system with three testing modes operates with a smart change of statuses: a correct answer advances the word, while an incorrect one returns it for review. After each answer, regardless of the result, GPT generates 3 example sentences with this word and its breakdown - the word is reinforced through context, not rote memorization. New → Semi-new → Learned. Words with the status "Learned" are no longer shown, and mistakes are highlighted in a separate mode and processed separately.
Result for the business:
The user focuses only on those words that truly need attention, does not waste time on what they already know, and does not miss those moments where mistakes are made. The system automatically adapts to each user. The bot is easily customizable for any language - just change the prompt.
Numbers:
- 3 testing modes: new words, mistakes, repetitions
- 4 word statuses: New → Semi-new → Learned / Mistake
- Customizable for any language without changing the architecture
Stack:
n8n · Telegram Bot API · GPT-4o Mini · Supabase · JavaScript Code Nodes