app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import pipeline
from typing import List

app = FastAPI(title="NER + Emotion API")

# ---------------------------------------------------------
# LOAD NER FIRST (PRIORITY LOAD)
# ---------------------------------------------------------
print("Loading NER model...")
ner_pipeline = pipeline(
    "ner",
    model="dslim/bert-base-NER",
    aggregation_strategy="simple"
)
print("NER model loaded.")

# ---------------------------------------------------------
# LOAD SENTIMENT SECOND
# ---------------------------------------------------------
print("Loading Sentiment model...")
sentiment_pipeline = pipeline(
    "text-classification",
    model="j-hartmann/emotion-english-distilroberta-base",
    top_k=1
)
print("Sentiment model loaded.")

# ---------------------------------------------------------
# REQUEST MODELS
# ---------------------------------------------------------
class TextInput(BaseModel):
    text: str


class SentimentInput(BaseModel):
    sentences: List[str]


# ---------------------------------------------------------
# HEALTH CHECK
# ---------------------------------------------------------
@app.get("/")
def home():
    return {"message": "NER + Emotion API is running"}


# ---------------------------------------------------------
# NER ENDPOINT
# ---------------------------------------------------------
# ---------------------------------------------------------
# NER ENDPOINT (UPDATED)
# ---------------------------------------------------------
@app.post("/analyze/ner")
def analyze_ner(data: TextInput):
    try:
        # REMOVED truncation=True to fix the 500 error
        results = ner_pipeline(data.text, aggregation_strategy="simple")
        
        persons = []
        locations = []
        organizations = []

        for entity in results:
            label = entity["entity_group"]
            word = entity["word"].strip()
            
            # dslim/bert-base-NER uses these labels:
            if label == "PER":
                persons.append(word)
            elif label == "LOC":
                locations.append(word)
            elif label == "ORG":
                organizations.append(word)

        return {
            "persons": list(set(persons)),
            "locations": list(set(locations)),
            "organizations": list(set(organizations))
        }
    except Exception as e:
        # This will help you see the exact error in HF logs
        print(f"Internal Error: {str(e)}")
        raise HTTPException(status_code=500, detail=str(e))


# ---------------------------------------------------------
# SENTIMENT ENDPOINT
# ---------------------------------------------------------
@app.post("/analyze/sentiment")
def analyze_sentiment(data: SentimentInput):
    try:
        results = sentiment_pipeline(
            data.sentences,
            truncation=True,
            max_length=512
        )

        processed_results = []

        for res_list in results:
            top_result = res_list[0]
            label = top_result["label"]
            score = top_result["score"]

            if label == "joy":
                polarity = score
            elif label in ["anger", "disgust", "fear", "sadness"]:
                polarity = -score
            else:
                polarity = 0.0

            processed_results.append({
                "label": label,
                "confidence": score,
                "polarity": polarity
            })

        return {"results": processed_results}

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))


if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)