feat: text editor and blog post

api/salience/salience.py

@@ -83,16 +83,31 @@ def text_rank(sentences, model_name='all-mpnet-base-v2'):
     adjacency[adjacency < 0] = 0
     return normalized_adjacency(adjacency)
 
+def extract(source_text, model_name='all-mpnet-base-v2'):
+    """
+    Main API function that extracts sentence positions and computes normalized adjacency matrix.
+
+    Returns:
+        sentence_ranges: List of (start, end) tuples for each sentence's character position
+        adjacency: (N × N) normalized adjacency matrix where N is the number of sentences.
+                   Each entry (i,j) represents the normalized similarity between sentences i and j.
+                   This matrix is returned to the frontend, which raises it to a power and computes
+                   the final salience scores via random walk simulation.
+    """
+    sentences, sentence_ranges = get_sentences(source_text)
+    adjacency = text_rank(sentences, model_name)
+    return sentence_ranges, adjacency
+
+
+# =============================================================================
+# Unused/Debugging Code
+# =============================================================================
+
 def terminal_distr(adjacency, initial=None):
     sample = initial if initial is not None else torch.full((adjacency.shape[0],), 1.)
     scores = sample.matmul(torch.matrix_power(adjacency, 10)).numpy().tolist()
     return scores
 
-def extract(source_text, model_name='all-mpnet-base-v2'):
-    sentences, sentence_ranges = get_sentences(source_text)
-    adjacency = text_rank(sentences, model_name)
-    return sentence_ranges, adjacency
-
 def get_results(sentences, adjacency):
     scores = terminal_distr(adjacency)
     for score, sentence in sorted(zip(scores, sentences), key=lambda xs: xs[0]):