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
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376 | class Scan(Media):
"""
Coordinates the semantic analysis of a submitted document, converting it into engrams and metadata
by orchestrating image extraction, LLM-driven summarization, and structured information parsing.
This class performs the following operations:
- Loads a PDF document and converts each page into Base64-encoded images.
- Generates an initial summary using a few preview pages.
- Scans each page with an LLM plugin to extract semantic content.
- Parses scan results into Engrams and assembles a full summary.
- Constructs a Meta object and emits an observation to the system.
Attributes:
id (str): Unique identifier for the scan session.
service (SenseService): Reference to the parent service orchestrating the scan.
page_images (list[str]): Base64-encoded representations of PDF pages.
sense_initial_summary (Plugin): Plugin used to generate an initial summary from early pages.
repo_id (str): Repository identifier for the document being scanned.
repo_ids (list[str] | None): List of repository IDs or None if no repository is specified.
document (Document | None): The document being processed.
tracking_id (str): ID for tracking the scan process.
Constants:
DPI (int): Resolution used for image extraction.
DPI_DIVISOR (int): Used to calculate zoom level for rendering.
TEST_PAGE_LIMIT (int): Max number of pages scanned during test runs.
MAX_CHUNK_SIZE (int): Max text length before recursive engram chunking.
SHORT_SUMMARY_PAGE_COUNT (int): Number of pages used to generate initial summary.
MAX_DEPTH (int): Maximum recursion depth for engram processing.
SECTION, H1, H3 (int): Enum-like values to manage tag-based engram splitting depth.
Methods:
parse_media_resource(document): Loads and validates a PDF, then initiates conversion.
_convert_pages_to_images(pdf, start_page, end_page): Converts PDF pages to images asynchronously.
_page_to_image(pdf, page_number): Converts and encodes a single PDF page to Base64.
_on_pages_converted(future): Handles post-conversion logic, triggering initial summary.
_generate_short_summary(): Sends preview pages to the LLM for initial semantic scan.
_on_short_summary(future): Kicks off page-by-page scanning after initial summary.
_scan_page(page_num): Sends a single page to the LLM plugin and extracts structured data.
_on_pages_scanned(future): Extracts structured context and initiates full summary.
_process_engrams(text_in, context, depth): Recursively splits and constructs Engrams from HTML text.
_generate_full_summary(summary): Uses full document content to generate final semantic summary.
_on_generate_full_summary(future): Wraps summary and Engrams into a Meta and emits final Observation.
"""
DPI = 72
DPI_DIVISOR = 72
TEST_PAGE_LIMIT = 100
MAX_CHUNK_SIZE = 1200
SHORT_SUMMARY_PAGE_COUNT = 4
MAX_DEPTH = 3
SECTION = 0
H1 = 1
H3 = 2
def __init__(self, parent_service: SenseService, repo_id: str, tracking_id: str):
self.scan_id = str(uuid.uuid4())
self.observation_id = str(uuid.uuid4())
self.repo_id: str = repo_id
if repo_id:
self.repo_ids: list[str] | None = [repo_id]
else:
self.repo_ids = None
self.service = parent_service
self.document: Document | None = None
self.tracking_id = tracking_id
self.page_images: list[str] = []
self.sense_initial_summary = self.service.sense_initial_summary
def parse_media_resource(self, document: Document) -> None:
async def send_message(observation_id: str, document_id: str) -> None:
self.service.send_message_async(
Service.Topic.OBSERVATION_CREATED, {'id': observation_id, 'parent_id': document_id}
)
self.service.run_task(send_message(self.observation_id, document.id))
logging.info('Parsing document: %s', document.file_name)
file_path: Traversable | Path
if document.root_directory == Document.Root.RESOURCE.value:
file_path = files(document.file_path).joinpath(document.file_name)
elif document.root_directory == Document.Root.DATA.value:
repo_root = os.getenv('REPO_ROOT')
if repo_root is None:
error = "Environment variable 'REPO_ROOT' is not set."
raise RuntimeError(error)
expanded_path = Path(repo_root + document.file_path).expanduser()
file_path = expanded_path / document.file_name
self.document = document
self.source_ids = [document.id]
try:
with file_path.open('rb') as file_ptr:
pdf_document: fitz.Document = fitz.open(stream=file_ptr.read(), filetype='pdf')
total_pages = pdf_document.page_count
if total_pages == 0:
error = 'PDF loaded with zero page count.'
raise RuntimeError(error)
self.page_images = [''] * total_pages
self.total_pages = total_pages
self._convert_pages_to_images(pdf_document, 0, total_pages)
except FileNotFoundError as e:
error = f'File {document.file_name} failed to open. {e}'
logging.exception(error)
except RuntimeError as e:
error = f'File {document.file_name} failed to open. {e}'
logging.exception(error)
def _convert_pages_to_images(self, pdf: fitz.Document, start_page: int, end_page: int) -> None:
coroutines = [self._page_to_image(pdf, i) for i in range(start_page, end_page)]
future = self.service.run_tasks(coroutines)
future.add_done_callback(self._on_pages_converted)
async def _page_to_image(self, pdf: fitz.Document, page_number: int) -> bool:
page = pdf.load_page(page_number)
zoom = Scan.DPI / Scan.DPI_DIVISOR
matrix = fitz.Matrix(zoom, zoom)
pix = page.get_pixmap(matrix=matrix)
# Convert the pixmap to PNG bytes in memory
img_bytes = pix.tobytes('png')
# Encode to Base64
encoded_img = base64.b64encode(img_bytes).decode('utf-8')
# Store the Base64 string in image_array
self.page_images[page_number] = encoded_img
return True
def _on_pages_converted(self, future: Future[Any]) -> None:
ret_functions = future.result()
del ret_functions
summary_future = self.service.run_task(self._generate_short_summary())
summary_future.add_done_callback(self._on_short_summary)
async def _generate_short_summary(self) -> Any:
plugin = self.sense_initial_summary
summary_images = self.page_images[: Scan.SHORT_SUMMARY_PAGE_COUNT]
if self.document is None:
error = 'Document must be set before generating prompt.'
raise ValueError(error)
prompt = PromptGenMeta(input_data={'file_path': self.document.file_path, 'file_name': self.document.file_name})
structured_response = {
'file_path': str,
'file_name': str,
'subject': str,
'audience': str,
'document_title': str,
'document_format': str,
'document_type': str,
'toc': str,
'summary_initial': str,
'author': str,
'date': str,
'version': str,
}
ret = self.sense_initial_summary['func'].submit(
prompt=prompt,
images=summary_images,
structured_schema=structured_response,
args=self.service.host.mock_update_args(plugin),
)
self.service.host.update_mock_data(plugin, ret)
initial_scan = json.loads(ret[0]['llm_response'])
return initial_scan
def _on_short_summary(self, future: Future[Any]) -> None:
result = future.result()
self.inital_scan = result
self.total_pages = min(Scan.TEST_PAGE_LIMIT, self.total_pages)
coroutines = [self._scan_page(i) for i in range(self.total_pages)]
future = self.service.run_tasks(coroutines)
future.add_done_callback(self._on_pages_scanned)
async def _scan_page(self, page_num: int) -> Any:
# logging.info(f"Scan Page: {page_num}")
plugin = self.service.sense_scan_page
initial_scan_copy = copy.copy(self.inital_scan)
initial_scan_copy.update({'page_number': page_num + 1})
prompt_scan = PromptScanPage(input_data=initial_scan_copy)
image = self.page_images[page_num]
ret = await asyncio.to_thread(
self.sense_initial_summary['func'].submit,
prompt=prompt_scan,
images=[image],
structured_schema=None,
args=self.service.host.mock_update_args(plugin),
)
self.service.host.update_mock_data(plugin, ret)
return ret[0]['llm_response']
def _on_pages_scanned(self, future: Future[Any]) -> None:
result = future.result()
self.meta_id = str(uuid.uuid4())
context: dict[str, str] = {}
context = copy.copy(self.inital_scan)
del context['summary_initial']
del context['toc']
self.engrams: list[Engram] = []
assembled = ''
for page in result['_scan_page']:
assembled += page
# matches1 = re.findall(r'<h1[^>]*>(.*?)</h1>', assembled, re.DOTALL | re.IGNORECASE)
# matches2 = re.findall(r'<h3[^>]*>(.*?)</h3>', assembled, re.DOTALL | re.IGNORECASE)
self._process_engrams(assembled, context)
future = self.service.run_task(self._generate_full_summary(assembled))
future.add_done_callback(self._on_generate_full_summary)
def _process_engrams(self, text_in: str, context: dict[str, str], depth: int = 0) -> None:
if len(text_in) > Scan.MAX_CHUNK_SIZE and depth < Scan.MAX_DEPTH:
tag = ''
if depth == Scan.SECTION:
tag = 'section'
if depth == Scan.H1:
tag = 'h1'
if depth == Scan.H3:
tag = 'h3'
depth += 1
pattern = rf'(?=<{tag}[^>]*>)'
parts = re.split(pattern, text_in, flags=re.IGNORECASE)
clean_parts = [part.strip() for part in parts if part.strip()]
for part in clean_parts:
match = re.search(rf'<{tag}[^>]*>(.*?)</{tag}>', part, re.DOTALL | re.IGNORECASE)
context_copy = context
if match:
tag_str = match.group(1).strip()
context.update({tag: tag_str})
context_copy = copy.copy(context)
self._process_engrams(part, context_copy, depth)
else:
if self.document is None:
error = 'Document id is None but a value is expected.'
raise RuntimeError(error)
engram = Engram(
str(uuid.uuid4()),
[self.inital_scan['file_path']],
[self.document.id],
text_in,
True,
context,
None,
[self.meta_id],
self.repo_ids,
None, # accuracy
None, # relevancy
int(datetime.now(timezone.utc).timestamp()),
)
self.engrams.append(engram)
async def _generate_full_summary(self, summary: str) -> Any:
plugin = self.service.sense_full_summary
initial_scan_copy = copy.copy(self.inital_scan)
initial_scan_copy.update({'full_text': summary})
prompt = PromptGenFullSummary(input_data=initial_scan_copy)
structure = {'summary_full': str, 'keywords': str}
ret = self.service.sense_full_summary['func'].submit(
prompt=prompt, images=None, structured_schema=structure, args=self.service.host.mock_update_args(plugin)
)
self.service.host.update_mock_data(plugin, ret)
llm_response = ret[0]['llm_response']
return llm_response
def _on_generate_full_summary(self, future: Future[Any]) -> None:
results = json.loads(future.result())
if self.document is None:
error = 'Document is None but expected not to be.'
raise RuntimeError(error)
meta = Meta(
self.meta_id,
Meta.SourceType.DOCUMENT.value,
[self.inital_scan['file_path'] + self.inital_scan['file_name']],
self.source_ids,
results['keywords'].split(','),
self.repo_ids,
self.inital_scan['summary_initial'],
Index(results['summary_full']),
self.document.id,
)
observation = Observation(
self.observation_id,
meta,
self.engrams,
int(datetime.now(timezone.utc).timestamp()),
self.document.id,
self.tracking_id,
)
self.service.host.update_mock_data_output(self.service, asdict(observation))
self.service.send_message_async(Service.Topic.OBSERVATION_COMPLETE, asdict(observation))
self.service.send_message_async(Service.Topic.DOCUMENT_COMPLETE, asdict(self.document))
|