1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332"""Markdown cleanup utilities using mdformat and pyspellchecker.
This module cleans up Docling markdown output before LLM processing.
Auto-fixes (via mdformat):
- Removes trailing whitespace
- Collapses multiple blank lines to one
- Ensures file ends with single newline
- Normalizes list markers and code block formatting
Spell checking (via pyspellchecker):
- Flags potential misspellings for LLM to verify against page image
- Skips code blocks, inline code, and URLs
- Uses technical dictionary (config/technical_dictionary.txt) to avoid
flagging domain-specific terms
"""
import logging
import re
from dataclasses import dataclass
from pathlib import Path
import mdformat
from spellchecker import SpellChecker
logger = logging.getLogger(__name__)
# Path to technical dictionary file
TECHNICAL_DICT_PATH = Path("config/technical_dictionary.txt")
@dataclass
class SpellingFlag:
"""A flagged potential misspelling for LLM review."""
word: str
suggestions: list[str]
line_number: int
context: str
@dataclass
class MarkdownCleanupResult:
"""Result of markdown cleanup processing."""
cleaned_text: str
formatting_changes: int # Number of characters changed by mdformat
spelling_flags: list[SpellingFlag]
words_checked: int
words_flagged: int
def _load_technical_dictionary() -> set[str]:
"""Load technical terms that should not be flagged as misspellings.
Returns:
Set of technical terms (lowercase)
"""
if not TECHNICAL_DICT_PATH.exists():
logger.warning(
f"Technical dictionary not found at {TECHNICAL_DICT_PATH}, "
"using empty dictionary"
)
return set()
try:
with open(TECHNICAL_DICT_PATH) as f:
terms = set()
for line in f:
# Skip comments and empty lines
line = line.strip()
if line and not line.startswith("#"):
terms.add(line.lower())
logger.debug(f"Loaded {len(terms)} technical terms from dictionary")
return terms
except Exception as e:
logger.error(f"Error loading technical dictionary: {e}")
return set()
def _extract_prose_words(text: str) -> list[tuple[str, int, str]]:
"""Extract words from prose sections, skipping code and URLs.
Args:
text: Markdown text
Returns:
List of (word, line_number, context) tuples
"""
words = []
lines = text.split("\n")
in_code_block = False
for line_num, line in enumerate(lines, 1):
# Track fenced code blocks
if line.strip().startswith("```"):
in_code_block = not in_code_block
continue
if in_code_block:
continue
# Skip indented code blocks (4 spaces or tab at start of line)
if line.startswith((" ", "\t")):
continue
# Remove inline code spans
line_clean = re.sub(r"`[^`]+`", "", line)
# Remove URLs
line_clean = re.sub(r"https?://[^\s\)\]]+", "", line_clean)
line_clean = re.sub(r"<[^>]+>", "", line_clean) # Email/URL in brackets
# Remove markdown link URLs but keep link text
line_clean = re.sub(r"\[([^\]]+)\]\([^\)]+\)", r"\1", line_clean)
# Remove image syntax
line_clean = re.sub(r"!\[([^\]]*)\]\([^\)]+\)", "", line_clean)
# Remove markdown formatting characters
line_clean = re.sub(r"[#*_\[\](){}|>~]", " ", line_clean)
# Extract words (letters only, min 2 chars)
word_matches = re.findall(r"\b[a-zA-Z]{2,}\b", line_clean)
for word in word_matches:
# Get context (surrounding text)
context = line.strip()[:80]
words.append((word, line_num, context))
return words
def _create_spell_checker(technical_terms: set[str]) -> SpellChecker:
"""Create and configure spell checker with technical dictionary.
Args:
technical_terms: Set of technical terms to add to dictionary
Returns:
Configured SpellChecker instance
"""
spell = SpellChecker(language="en", distance=1)
# Add technical terms to the dictionary
if technical_terms:
spell.word_frequency.load_words(list(technical_terms))
return spell
def format_markdown(text: str) -> tuple[str, int]:
"""Apply mdformat to fix markdown formatting issues.
Args:
text: Raw markdown text
Returns:
Tuple of (formatted_text, characters_changed)
"""
original_len = len(text)
try:
# mdformat.text() returns formatted markdown
# Extensions like mdformat-gfm are auto-detected if installed
formatted = mdformat.text(text)
chars_changed = abs(len(formatted) - original_len)
if chars_changed > 0:
logger.info(
f"mdformat applied: {chars_changed} characters changed "
f"({original_len} -> {len(formatted)})"
)
return formatted, chars_changed
except Exception as e:
logger.warning(f"mdformat failed, returning original text: {e}")
return text, 0
def check_spelling(
text: str,
technical_terms: set[str] | None = None,
max_flags: int = 50,
) -> tuple[list[SpellingFlag], int, int]:
"""Check spelling and return flagged words for LLM review.
Does NOT auto-correct - only flags potential issues for the LLM
to verify against the visual page image.
Args:
text: Markdown text to check
technical_terms: Optional set of technical terms to ignore
max_flags: Maximum number of flags to return (avoid overwhelming LLM)
Returns:
Tuple of (spelling_flags, words_checked, words_flagged)
"""
if technical_terms is None:
technical_terms = _load_technical_dictionary()
spell = _create_spell_checker(technical_terms)
# Extract prose words with line numbers and context
word_data = _extract_prose_words(text)
words_checked = len(word_data)
# Find misspelled words
unique_words = {w.lower() for w, _, _ in word_data}
misspelled = spell.unknown(unique_words)
# Filter out likely proper nouns (capitalized words)
# and very short words that are often acronyms
flags: list[SpellingFlag] = []
seen_words: set[str] = set()
for word, line_num, context in word_data:
word_lower = word.lower()
# Skip if already flagged or not misspelled
if word_lower in seen_words or word_lower not in misspelled:
continue
# Skip likely proper nouns (first letter capitalized, not start of sentence)
# This is a heuristic - LLM will verify against image
if word[0].isupper() and not context.startswith(word):
continue
# Skip very short words (likely acronyms)
if len(word) < 3:
continue
# Get suggestions
suggestions = list(spell.candidates(word_lower) or [])[:3]
# Only flag if we have suggestions (otherwise likely a valid rare word)
if suggestions:
flags.append(
SpellingFlag(
word=word,
suggestions=suggestions,
line_number=line_num,
context=context,
)
)
seen_words.add(word_lower)
if len(flags) >= max_flags:
logger.info(f"Reached max spelling flags ({max_flags}), stopping")
break
logger.info(
f"Spell check complete: {words_checked} words checked, "
f"{len(flags)} potential misspellings flagged"
)
return flags, words_checked, len(flags)
def cleanup_markdown(
text: str,
enable_formatting: bool = True,
enable_spell_check: bool = True,
technical_terms: set[str] | None = None,
) -> MarkdownCleanupResult:
"""Apply markdown formatting fixes and spell checking.
This is the main entry point for the cleanup pipeline.
Args:
text: Raw markdown text from Docling
enable_formatting: Whether to apply mdformat fixes
enable_spell_check: Whether to check spelling
technical_terms: Optional technical dictionary (loads from file if None)
Returns:
MarkdownCleanupResult with cleaned text and flags
"""
cleaned_text = text
formatting_changes = 0
spelling_flags: list[SpellingFlag] = []
words_checked = 0
words_flagged = 0
# Step 1: Apply mdformat for deterministic fixes
if enable_formatting:
cleaned_text, formatting_changes = format_markdown(text)
# Step 2: Check spelling on the formatted text
if enable_spell_check:
if technical_terms is None:
technical_terms = _load_technical_dictionary()
spelling_flags, words_checked, words_flagged = check_spelling(
cleaned_text, technical_terms
)
return MarkdownCleanupResult(
cleaned_text=cleaned_text,
formatting_changes=formatting_changes,
spelling_flags=spelling_flags,
words_checked=words_checked,
words_flagged=words_flagged,
)
def format_spelling_flags_for_prompt(flags: list[SpellingFlag]) -> str:
"""Format spelling flags as text for inclusion in LLM prompt.
Args:
flags: List of SpellingFlag objects
Returns:
Formatted string for prompt, or empty string if no flags
"""
if not flags:
return ""
lines = ["**Flagged potential misspellings** (verify against page image):"]
for flag in flags:
suggestions_str = ", ".join(f'"{s}"' for s in flag.suggestions)
lines.append(
f"- Line {flag.line_number}: \"{flag.word}\" - "
f"suggestions: [{suggestions_str}]"
)
return "\n".join(lines)