Sentiment Analysis at Scale: NLP Pipelines with Hugging Face and NestJS

Understanding customer sentiment helps brands react to crises before they escalate. Here is how we built a real-time sentiment classification pipeline.

1. Model Selection

We use the distilbert-base-uncased-fined-tuned-sst-2-english model from Hugging Face for fast inference:

python
from transformers import pipeline
classifier = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english")
result = classifier("This product is amazing!")
# [{'label': 'POSITIVE', 'score': 0.9998}]

2. NestJS Microservice Gateway

The NestJS backend queues incoming social media mentions and dispatches them to the Python inference service:

typescript
@Injectable()
export class SentimentService {
  async classify(text: string): Promise<SentimentResult> {
    const response = await this.httpService.post('http://ml-service:5000/predict', { text });
    return response.data;
  }
}

3. Real-Time Feed with Socket.IO

Classified results are pushed to connected dashboards via Socket.IO, showing live sentiment scores color-coded as green (positive), yellow (neutral), or red (negative).

Summary

By separating the ML inference layer from the API gateway, the system scales independently and processes thousands of mentions per minute.