from fastapi import FastAPI
from pydantic import BaseModel
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import dateparser
from datetime import datetime
import re

app = FastAPI()

# Load classification model
classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")

# Load summarization model
summarizer_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-small")
summarizer_model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-small")

# Labels
labels = ["task", "event", "reminder", "meeting", "relationship", "note", "journal", "memory", "other"]

class TextInput(BaseModel):
    text: str

def extract_dates(text):
    time_expressions = re.findall(
        r'\b(kal|aaj|parso|raat|subah|shaam|dopahar|[0-9]{1,2} baje|next week|tomorrow|today|yesterday|Monday|Tuesday|Wednesday|Thursday|Friday|Saturday|Sunday|[\d]{1,2}/[\d]{1,2}/[\d]{2,4})\b',
        text, flags=re.IGNORECASE)
    parsed = [str(dateparser.parse(t)) for t in time_expressions if dateparser.parse(t)]
    return list(set(parsed)), list(set(time_expressions))

def detect_tense(parsed_dates):
    now = datetime.now()
    tenses = set()
    for d in parsed_dates:
        dt = dateparser.parse(d)
        if not dt:
            continue
        if dt < now:
            tenses.add("past")
        elif dt > now:
            tenses.add("future")
        else:
            tenses.add("present")
    return list(tenses) if tenses else ["unknown"]

def generate_summary(text):
    input_ids = summarizer_tokenizer("summarize: " + text, return_tensors="pt").input_ids
    output_ids = summarizer_model.generate(input_ids, max_length=50, num_beams=4, early_stopping=True)
    return summarizer_tokenizer.decode(output_ids[0], skip_special_tokens=True)

@app.post("/analyze")
async def analyze(input: TextInput):
    text = input.text
    classification = classifier(text, labels)
    best_label = classification['labels'][0]
    scores = dict(zip(classification['labels'], classification['scores']))
    parsed_dates, time_mentions = extract_dates(text)
    tenses = detect_tense(parsed_dates)
    summary = generate_summary(text)

    return {
        "type": best_label,
        "confidence_scores": scores,
        "time_mentions": time_mentions,
        "parsed_dates": parsed_dates,
        "tense": tenses,
        "summary": summary
    }