---
license: mit
language: en
tags:
- rnn
- lstm
- chatbot
- dialogue-generation
- funny
pretty_name: "RNN AI-bot - The Apologetic Seat Obsessor"
---

# RNN AI-bot

This is a small, simple, and **grammatically flawless** but semantically clueless chatbot. It was trained on the DailyDialog dataset using a character-level RNN (LSTM) with a vocabulary of 10,002 tokens (including PAD and UNK).

## Model Details

- **Architecture**: LSTM (256 units) + Embedding (128 dims)
- **Input Length**: 50 tokens (context window)
- **Vocabulary Size**: 10,002
- **Training Data**: DailyDialog (13k dialogues, ~8 turns each)
- **Framework**: TensorFlow/Keras

## How to Use

```python
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.sequence import pad_sequences

model = load_model("best_model.h5")

# You'll need the word list and ID mappings.
# See the GitHub repo for full chat script.Performance & Quirks
The model generates grammatically correct English sentences but has no semantic understanding. It tends to:

Apologize excessively ("I am sorry.")

Obsess over the word "seat"

Insert "?" randomly

Get stuck in repetitive loops

Example Conversation
text
User: How are you?
Bot:  I am a student in the mood for a while .
User: What are you studying?
Bot:  I am applying for a vacation .
User: Did you get the vacation?
Bot:  Yes , I am going to get a lot of the ? .
User: Where will you go?
Bot:  I don't know what's a ? .
User: Seat?
Bot:  Yes , I am going to get a lot of the ? .
User: ?
Bot:  I don't know what's a seat .
Why Does It Behave Like This?
Small training: Only 10 epochs on CPU.

Limited context: 50-token window.

No pretrained embeddings: Learned everything from scratch.

Data bias: The word "seat" appears frequently in DailyDialog.

Full Source Code
All preprocessing, training, and chat scripts are available on GitHub:

👉 zyxdtt/Fun-Project/tree/main/RNN_AI-bot

There you'll find:

preprocessing.cpp – C++ data pipeline

RNN_train.py – Model training code

chat_test.py – Interactive chat script

Chat screenshots showing its hilarious responses

Fun Fact
This is the first AI project completed by a university freshman. It may be dumb, but it's 100% my own work — from data processing to deployment.

This bot proves that perfect grammar doesn't equal intelligence. It's a lesson in overfitting, data bias, and the importance of semantics. 😅