Language Coach
Minimal language coach with conversation memory.
Summary
A minimal language-coach chatbot where the user picks between translating sentences or having grammar corrected via a dropdown. The bot remembers conversation history through BufferMemory.
Stack
Highlights
- Two modes - translation and grammar correction
- BufferMemory for conversational context
- Local LLM via Ollama
Background
AI course project. A minimal language-coach chatbot - focus on conversation context and prompt steering, not on massive feature scope.
What it is
The user picks between two modes via a dropdown:
- Translation - write a sentence, get it translated
- Grammar correction - write a sentence, get it corrected with an explanation
The bot remembers conversation history through LangChain's BufferMemory, so follow-ups like "continue in the same style" or "what did I just write?" work.
Technical choices
- LangChain.js - for chain construction and memory management
- Ollama - local LLM, no external API dependency
- React - minimal UI: dropdown + text field + message log
What I focused on
Keeping the scope small and doing the few things really well. No multi-language detection, no user accounts, no history between sessions. Just the dropdown, the field, and a bot that actually remembers what you said last time.
What I took away
BufferMemory is one of the simpler LangChain components, but it becomes powerful paired with a tight scope. When the bot has a clear job (translate or correct) short conversations feel natural without me building complex state management.