DeFraudAI

def predict_deepfake(image_data: bytes) -> dict:
    # Decode, resize to model input, normalise to [0, 1]
    img = Image.open(io.BytesIO(image_data)).resize((224, 224))
    img_array = np.array(img) / 255.0
    img_array = np.expand_dims(img_array, axis=0)   # add batch dim

    prediction = model.predict(img_array)[0][0]
    return {
        "is_deepfake": bool(prediction > 0.5),
        "confidence": float(prediction),  # 0 = real, 1 = deepfake
    }

app.post('/api/analyze', zValidator('json', AnalysisRequestSchema), async (c) => {
  const { content } = c.req.valid('json');

  // Deterministic hash - same content always maps to the same cache key
  const data = new TextEncoder().encode(content.toLowerCase().trim());
  const hash = await crypto.subtle.digest('SHA-256', data);
  const contentHash = [...new Uint8Array(hash)]
    .map(b => b.toString(16).padStart(2, '0')).join('');

  // D1 cache lookup - skip Gemini on repeat submissions
  const cached = await c.env.DB.prepare(
    'SELECT * FROM analysis_results WHERE content_hash = ? LIMIT 1'
  ).bind(contentHash).first();
  if (cached) return c.json(JSON.parse(String(cached.result)));

  const result = await analyzeContent(content);   // → Gemini API
  await c.env.DB.prepare(
    'INSERT INTO analysis_results (content_hash, ...) VALUES (?, ...)'
  ).bind(contentHash /*, ... */).run();
  return c.json(result);
});