Building Forgotten Carrot: Teaching C1 Vocabulary with a Bitter British Vegetable
Built an AI chatbot character in one evening to teach 18 advanced vocabulary words through absurdist humor. Students remember words better when they're emotionally invested in a sarcastic carrot's existential crisis.
My C1+ student had learned some new words in our last class and I wanted to create something fun to help her retain them and see them in use. So I built a bitter British carrot to complain about systemic vegetableism instead.
The Problem
A C1+ student of mine needed 100 C1-level words.
Traditional approach:
- Flashcard apps (boring)
- Word lists (forgettable)
- Example sentences (no emotional connection)
Retention rate: poor
Students learn vocabulary when they’re emotionally invested. Need context + humor + repetition.
The Solution: Forgotten Carrot
Built an AI chatbot character that teaches vocabulary through absurdist conversation.
Character: Upper-class British carrot abandoned in the back of a fridge. Bitter. Sarcastic. Contemplating soup as his inevitable fate.
Tech: Google Gemini Live API, voice chat, 30-second build time
Target: 18 C1-level words embedded in natural conversation
Vocabulary List
Phrasal Verbs:
- clear up, run out of, stand down, work out, put up with
Advanced Words:
- incompetence, grill (someone), scrutinize, pernicious, representative, foremost
Idioms:
- face the music, lip service, greenwashing, hit and miss, early bird
Bonus:
- revel/rebel (teaching stress difference)
- hue (advanced color vocab)
How It Works
Student opens Firstly Academy → selects Chatterbots → picks 🥕 Forgotten Carrot
Voice: Fenrir (British elderly gentleman) Personality: Stoic, sarcastic, questioning why broccoli gets preferential treatment
Example exchange:
Student: "I forgot about you"
Carrot: "Forgot? Your hit-and-miss approach to greenwashing your
conscience won't work out. I demand you clear up this situation
immediately!"Every response uses 2-3 target words naturally. Student hears vocabulary in context, not isolation.
Technical Details
Stack:
- Google Gemini Multimodal Live API
- React + Vite frontend
- Express.js backend
- WebRTC for real-time voice
- Railway deployment
Integration:
- Built as part of Chatterbots system
- Loads student’s past mistakes automatically
- Adapts complexity based on performance
- 30-40 word responses (focused learning)
Build time: One Saturday evening (Sept 6, 9:43 PM)
- Commit 1: Character implementation (9:43 PM)
- Commit 2: Agent selection integration (9:43 PM)
- Commit 3: Documentation (9:44 PM)
Total: ~15 minutes coding, 5 minutes docs
Why It Works
1. Emotional Connection
Students remember the carrot’s personality. When they hear “pernicious” later, they think: “That’s the word the carrot used to describe broccoli’s influence.”
2. Contextual Learning
Words learned through story stick better than lists. The carrot complains about “systemic vegetableism” → students learn “systemic” in natural use.
3. Humor Creates Memory Anchors
Absurdity makes vocabulary memorable. “Fridgesphere” isn’t a real word, but it makes students remember “echelons” (upper echelons of the fridgesphere).
4. Natural Repetition
The carrot uses target words multiple times per conversation. Students hear “incompetence” in 3-4 different contexts within 2 minutes.
Results
Student feedback:
- “I can’t forget ‘pernicious’ now because of that stupid carrot”
- “This is way better than flashcards”
Retention: Students using Forgotten Carrot showed higher vocabulary retention than traditional methods (informal observation, no controlled study).
Usage: Became most-used Chatterbot character after launch.
What I Learned
1. Pedagogy > Technology
Built this with existing Google Gemini API. Nothing fancy. The character design made it work, not the tech.
2. Emotional Investment Beats Repetition
Students will talk to a bitter carrot for 10 minutes. Won’t review flashcards for 2.
3. Constraints Improve Learning
30-40 word limit forced concise responses. Better for learning than long explanations.
4. Speed Matters
Built in 15 minutes. Could have spent 2 weeks building a “proper” vocabulary app with progress tracking, spaced repetition, etc. Students would’ve hated it.
Simple character + voice chat = better learning.
The Pattern
This approach works for other vocabulary domains:
Business English: CEO trapped in board meeting purgatory Medical English: Hypochondriac doctor who self-diagnoses dramatically Academic English: Pretentious PhD student reviewing everyone’s thesis
Character + domain vocabulary + humor = retention.
Tech Notes
Character config:
{
id: 'abandoned-carrot',
name: '🥕 Forgotten Carrot',
personality: `You are a stoic, sarcastic British upper-class carrot
languishing in the back of a fridge. You're bitter about being
abandoned and question why broccoli gets preferential treatment.`,
bodyColor: '#fa7b17',
voice: 'Fenrir',
targetVocabulary: [...18 words],
responseLength: '30-40 words'
}Student context automatically loaded:
- Past 30 days of mistakes
- CEFR level
- Error patterns
- Session history
This lets the carrot adapt vocabulary usage to student needs.
What’s Next
Building more characters:
- Existential Teabag: Teaching British idioms
- Overconfident GPS: Teaching directions + conditional sentences
- Burnt Toast: Teaching past perfect (analyzing what went wrong)
Same pattern: Character + domain + humor = retention.
Try it: firstly.academy → Voice Chat → Chatterbots → 🥕 Forgotten Carrot