yogeshpandey586/chatbot-intent-classifier

ModernBERT Chatbot Intent Classifier

A fine-tuned ModernBERT model for enterprise chatbot intent classification. This model classifies user queries into predefined intent categories and is designed to be used as the first stage of a chatbot pipeline before Retrieval-Augmented Generation (RAG) or other downstream workflows.

Supported Intents

The model predicts one of the following intents:

• —Business
• —Greeting
• —SmallTalk
• —Complaint
• —Feedback
• —Thanks
• —Goodbye
• —OutOfDomain

Intended Use

This model is designed for:

• —Enterprise AI Chatbots
• —Customer Support Bots
• —HR Assistants
• —Knowledge Base Assistants
• —Intent Detection
• —Query Routing
• —RAG Pipeline Routing

Example flow:

User Query
      │
      ▼
Intent Classifier
      │
      ├── Business
      │      └── RAG Pipeline
      │
      ├── Greeting
      │      └── Greeting Response
      │
      ├── SmallTalk
      │      └── Small Talk Response
      │
      ├── Complaint
      │      └── Complaint Workflow
      │
      ├── Feedback
      │      └── Feedback Workflow
      │
      ├── Thanks
      │      └── Thank You Response
      │
      ├── Goodbye
      │      └── Goodbye Response
      │
      └── OutOfDomain
             └── Reject Politely

Base Model

• —Model: answerdotai/ModernBERT-base
• —Task: Sequence Classification
• —Framework: Hugging Face Transformers
• —Language: English

Example Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="yogeshpandey586/chatbot-intent-classifier"
)

print(classifier("Hello"))
print(classifier("What services do you provide?"))
print(classifier("Tell me a joke"))

Example Output

[
    {
        "label": "Greeting",
        "score": 0.998
    }
]

Training Dataset

The model was fine-tuned using a custom intent classification dataset containing manually labeled chatbot queries.

Each sample contains:

• —User Query
• —Intent Label

Example:

Training Details

• —Base Model: ModernBERT-base
• —Fine-tuning Task: Sequence Classification
• —Framework: Hugging Face Transformers
• —Optimizer: AdamW
• —Loss Function: Cross Entropy Loss

Evaluation

The model was evaluated using a validation dataset.

Evaluation Metric:

• —Accuracy

This model is intended for production chatbot routing where low latency and high intent classification accuracy are required.

Limitations

This model is intended only for intent classification.

It is not designed for:

• —Toxicity Detection
• —Sentiment Analysis
• —Emotion Detection
• —Translation
• —Summarization
• —Question Answering
• —Text Generation

Performance may decrease on:

• —Languages other than English
• —Ambiguous user queries
• —Multi-intent queries
• —Queries outside the training domain

Production Architecture

User Query
      │
      ▼
ModernBERT Intent Classifier
      │
      ▼
Intent Router
      │
      ├── Business → RAG
      ├── Greeting → Greeting Service
      ├── SmallTalk → Small Talk Service
      ├── Complaint → Complaint Service
      ├── Feedback → Feedback Service
      ├── Thanks → Thank You Service
      ├── Goodbye → Goodbye Service
      └── OutOfDomain → Reject Response

Technical Specifications

Developer

Yogesh Pandey

AI Developer

License

Apache-2.0

Citation

@misc{yogeshpandey2026intentclassifier,
  title={ModernBERT Chatbot Intent Classifier},
  author={Yogesh Pandey},
  year={2026},
  publisher={Hugging Face}
}