app.py

# app.py
import os
import math
import requests
from flask import Flask, request, jsonify
from flask_cors import CORS
from langdetect import detect

# -----------------------------------------------------------------------------
# Configuration
# -----------------------------------------------------------------------------
HF_API_URL = "https://api-inference.huggingface.co/models/YOUR_USERNAME/YOUR_MODEL"
HF_TOKEN = os.getenv("HF_TOKEN")

HEADERS = {
    "Authorization": f"Bearer {HF_TOKEN}",
    "Content-Type": "application/json"
}

app = Flask(__name__)
CORS(app)

# -----------------------------------------------------------------------------
# Utility Functions
# -----------------------------------------------------------------------------
def entropy(probs):
    """Shannon entropy as epistemic uncertainty indicator."""
    return -sum(p * math.log2(p) for p in probs if p > 0)

def normalize_labels(hf_output):
    """
    Normalize Hugging Face output into a stable schema.
    Expected HF format:
    [
      {"label": "HUMAN", "score": 0.73},
      {"label": "AI", "score": 0.27}
    ]
    """
    result = {item["label"].lower(): float(item["score"]) for item in hf_output}
    human_p = result.get("human", 0.0)
    ai_p = result.get("ai", 0.0)

    return human_p, ai_p

def hf_inference(text):
    payload = {"inputs": text}
    r = requests.post(HF_API_URL, headers=HEADERS, json=payload, timeout=30)
    r.raise_for_status()
    return r.json()

# -----------------------------------------------------------------------------
# Core Endpoint
# -----------------------------------------------------------------------------
@app.route("/analyze", methods=["POST"])
def analyze():
    data = request.get_json()
    text = data.get("text", "").strip()

    if not text:
        return jsonify({"error": "Empty input"}), 400

    # 1. Language detection (supports linguistic auditing)
    try:
        language = detect(text)
    except Exception:
        language = "unknown"

    # 2. Hugging Face inference
    hf_raw = hf_inference(text)

    if not isinstance(hf_raw, list):
        return jsonify({"error": "Unexpected model response", "raw": hf_raw}), 500

    human_p, ai_p = normalize_labels(hf_raw)

    # 3. Decision
    label = "Human" if human_p >= ai_p else "Machine"
    confidence = max(human_p, ai_p)

    # 4. Epistemic uncertainty
    H = entropy([human_p, ai_p])

    # 5. Explainability placeholder (XAI-ready schema)
    explainability_stub = {
        "method": "pending",
        "note": (
            "This model endpoint does not natively expose SHAP/LIME. "
            "Post-hoc explainability must be computed locally using a "
            "replicated model or proxy explainer."
        ),
        "token_attributions": []
    }

    # 6. Fairness metadata (for downstream auditing)
    fairness_context = {
        "language": language,
        "human_probability": human_p,
        "ai_probability": ai_p,
        "entropy": H
    }

    response = {
        "prediction": {
            "label": label,
            "confidence": round(confidence, 4)
        },
        "probabilities": {
            "human": round(human_p, 4),
            "machine": round(ai_p, 4)
        },
        "uncertainty": {
            "entropy": round(H, 4),
            "interpretation": (
                "High entropy indicates epistemic ambiguity; "
                "classification should be treated cautiously."
            )
        },
        "linguistic_context": {
            "detected_language": language
        },
        "explainability": explainability_stub,
        "fairness_audit_fields": fairness_context
    }

    return jsonify(response)

# -----------------------------------------------------------------------------
# Health Check
# -----------------------------------------------------------------------------
@app.route("/", methods=["GET"])
def index():
    return jsonify({
        "system": "HATA API",
        "capabilities": [
            "Human vs AI classification",
            "Probability calibration",
            "Uncertainty estimation",
            "Language-aware auditing",
            "Explainability-ready schema",
            "Fairness instrumentation"
        ]
    })

# -----------------------------------------------------------------------------
if __name__ == "__main__":
    app.run(host="0.0.0.0", port=5000, debug=True)