folder-classifier 0.3.4__py3-none-any.whl → 0.3.5__py3-none-any.whl

folder_classifier/app.py

@@ -17,7 +17,7 @@ class FolderClassifierAPI:
         logging.basicConfig(level=logging.INFO)
         self.logger = logging.getLogger(__name__)
         self.classifier = FolderClassifier(app_name=model_config.app_name, deployment=model_config.deployment, model=model_config.model)
-        self.logger.info(f"Successfully initialized Folder Classifier using config: {model_config}")
+        self.logger.info(f"Successfully initialized Folder Classifier API using config: {model_config}")
 
     @web_api.post("/predict")
     async def predict(self, request: FolderClassificationRequest) -> FolderClassificationResponse:

folder_classifier/classifier.py

@@ -1,28 +1,20 @@
 import json
 import logging
-import os
 from typing import Tuple, Dict, Any
 
 from ray import serve
-from ray.serve.handle import DeploymentHandle
 
 from folder_classifier.dto import FolderClassificationRequest, FolderClassification
 from folder_classifier.util import build_folder, render_tree
-from openai import AsyncOpenAI
-
-
-OPENAI_BASE_URL = os.getenv("OPENAI_BASE_URL", "https://internal-multimodal-services.dev.cortoaws.com/serve/qwen3-4b-classifier/v1")
-OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "DUMMY_KEY")
-MODEL = os.getenv("MODEL", "Qwen3-4B-Instruct-2507-classifier-FP8")
 
 SYSTEM_PROMPT = """
 You are a strict text classifier.
 Output MUST be a single JSON object with exactly two keys: "category" and "reasoning".
 - "category" ∈ {"matter","other"} (lowercase).
-- "reasoning" ≤ 30 words.
+- "reasoning" ≤ 30 words, citing rule IDs (e.g., RC+R7 or R1+R3) and up to two brief evidence tokens.
 - Do not include any double-quote (") characters inside the value of "reasoning"; use single quotes ' instead.
 - No backticks, no code fences, no extra text. Return one JSON object only.
-If unsure, choose "other".
+No step-by-step thinking. No extra keys. If unsure, choose "other".
 """.strip()
 
 USER_PROMPT_TEMPLATE = """
@@ -36,21 +28,28 @@ Definitions
   (c) legal-work indicators are absent, or
   (d) there are zero files with extensions anywhere in the tree, or
   (e) contents are exclusively non-legal domains (finance/accounting/IT/admin) with no legal-work indicators, or
-  (f) the ROOT NAME is generic/system-generated and not matter-specific (see R9).
+  (f) the ROOT NAME is non-matter-specific, including either:
+      (f1) system/provenance roots (e.g., “CORTO Generated”, “Exports”, “Uploads”, “Dropbox Shared”, “OneDrive”, “SharePoint”,
+           “Archive”, “Backup”, “Temp”, “Incoming”, “Outbox”, “Bulk Import”), or
+      (f2) informal/placeholder roots lacking client/company/matter cues (e.g., “hello”, “test”, “misc”, “docs”, “files”,
+           “images”, “photos”, “scans”, “new folder”, “untitled”, “random”; very short single tokens with no digits).
 
 Decision Rules (apply in order; case-insensitive)
-RC_container: Any indication the FOLDER TREE holds more than one distinct matter (e.g., multiple top-level matter-like subfolders; repeated separate matter numbers/clients) → category=other.
+RC_container: Any indication the FOLDER TREE holds more than one distinct matter → category=other.
+  • Signals include: ≥2 top-level subfolders whose names match matter-like patterns (e.g., “<number> - <id>”, “<client> - <topic>”), even if those subfolders are empty.
+  • Repetitive top-level “matter-number” patterns also indicate a container.
 R0_subfolder: ROOT NAME equals a common matter subfolder/stage/type (e.g., Correspondence/Emails, File Notes/Attendance Notes, Searches/Certificates, Court Documents/Pleadings/Evidence/Disclosure/Discovery, Drafts/Final/Signed/Executed, Billing/Invoices/Time Records) → category=other.
 R7_files_present: Must contain ≥1 file with an extension (e.g., .pdf, .docx) anywhere in the tree; if none → category=other.
-R8_nonlegal_only: If the tree shows strong non-legal domain signals (finance/accounting/IT/admin) and NO legal-work indicators (R2–R4 or R6), classify as other.
-    • Finance/Accounting examples: BAS/Business Activity Statement, IAS, GST, Tax Return, PAYG, Payroll, Timesheets, Payslips, Superannuation, Xero/MYOB/QuickBooks exports, General Ledger, Trial Balance, Journals, Bank Statements.
-    • IT/Systems examples: Backups, Logs, Source Code, Git, DevOps, Server/Network/VPN, Mailboxes, Google Workspace/Microsoft 365 admin.
-    • Admin-only examples: generic receipts, vendor invoices, expense folders without legal context.
-    • Note: “Billing/Invoices/Time Records” inside a matter is a typical legal subfolder; R8 applies only when legal indicators are entirely absent.
-R9_generic_rootname: If the ROOT NAME is generic/system-generated and not matter-specific → category=other (even if legal documents appear underneath).
-    • Examples: "CORTO Generated", "Generated Files", "Exports", "Uploads", "Scans", "Shared", "Dropbox Shared", "Google Drive", "OneDrive", "SharePoint", "Archive", "Backup", "Temp", "Incoming", "Outbox", "Bulk Import".
-    • Heuristic: words like generated/export/import/sync/shared/archive/backup/temp/uploads/downloads indicate a system or generic container, not a single matter.
-R1_rootname: ROOT NAME resembles a single matter/container (matter/file/case number; client/surname/company; or a combination such as “12345 Smith” or “Smith – Contract Dispute”).
+R8_nonlegal_only: If the tree shows strong non-legal domain signals and NO legal-work indicators (R2–R4 or R6), classify as other.
+  • Finance/Accounting examples: BAS/IAS/GST/Tax Return, PAYG, payroll/timesheets/payslips, superannuation, Xero/MYOB/QuickBooks exports, GL/Trial Balance/Journals, bank statements, invoices/receipts, reconciliations, aged AR/AP, debtor/creditor lists, debtor collection schedules, AR “demand/reminder” letters in collections context.
+  • IT/Systems examples: backups, logs, source code, Git/DevOps, server/network/VPN, mailboxes, workspace/365 admin.
+  • Admin-only examples: generic receipts/vendor invoices/expenses without legal context.
+R9_non_matter_specific_rootname: ROOT NAME is non-matter-specific → category=other.
+  • R9a_system_provenance: matches system/provenance roots (generated/export/import/shared/archive/backup/temp/uploads/
+                           downloads/platform names like Dropbox/OneDrive/SharePoint).
+  • R9b_placeholder_informal: matches informal/placeholder roots with no client/company/matter cues (single common words,
+                              “new folder”, “untitled”, short tokens without digits).
+R1_rootname: ROOT NAME resembles a single matter/container (matter/case/file number; client/surname/company; or a combination like “12345 Smith”, “Smith – Contract Dispute”, “Brown – Lease Review”).
 R2_initial_docs: Early-stage matter docs (cost agreement/disclosure, retainer/engagement, intake/onboarding).
 R3_legal_docs: Legal document types (agreement, contract, deed, will, affidavit, statement, advice, brief, pleadings, court forms, subpoena, orders, judgment, undertaking, notice of appeal, docket/case forms).
 R4_legal_subfolders: Typical legal subfolders (correspondence/emails, file notes/attendance notes, searches/certificates, court documents/evidence/disclosure/discovery, drafts/final/signed/executed, billing/invoices/time records).
@@ -63,16 +62,17 @@ Decision
 - Else if NOT R7_files_present → category=other (stop).
 - Else if R8_nonlegal_only → category=other (stop).
 - Else if R9_generic_rootname → category=other (stop).
-- Else if R1_rootname AND any of {R2_initial_docs, R3_legal_docs, R4_legal_subfolders, R6_jurisdiction} AND no multi-matter signal → category=matter.
-  (R5_support_filename_patterns cannot be used alone to justify "matter"; it is supportive only.)
+- Else if R10_weak_rootname → category=other (stop).
+- Else if R1_rootname AND any of {R2, R3, R4, R6} AND no multi-matter signal → category=matter.
+  (R5 is supportive only and cannot justify "matter" by itself.)
 - Else → category=other.
 
 Normalization
-- “File with extension”: name containing a period followed by a 1–5 char alphanumeric extension (e.g., .pdf, .docx). Ignore leading/trailing periods.
-- Treat hyphens/underscores as separators. Ignore file extensions for semantic matching beyond the presence test. Tolerate minor typos.
+- “File with extension”: period + 1–5 char alphanumeric extension (e.g., .pdf, .docx). Ignore leading/trailing periods.
+- Treat hyphens/underscores as separators. Ignore extensions for semantic matching beyond presence. Tolerate minor typos.
 
-Output format (JSON only; no prose before/after):
-{"category": "<matter|other>", "reasoning": "≤30 words citing R# and 1–2 evidence tokens>"}
+Output (JSON only; no prose before/after):
+{"category": "<matter|other>", "reasoning": "<≤30 words citing R# and 1–2 evidence tokens>"}
 
 FOLDER TREE:
 {folder_tree}
@@ -87,7 +87,7 @@ class FolderClassifier:
         self.logger = logging.getLogger(__name__)
         self.model_handle = serve.get_deployment_handle(app_name=app_name, deployment_name=deployment)
         self.model = model
-        self.logger.info(f"Successfully initialized FolderClassifier")
+        self.logger.info(f"Successfully initialized Folder Classifier with remote Ray model: {self.model}")
 
     async def predict(self, request: FolderClassificationRequest) -> Tuple[str, str]:
         content = ""
@@ -98,7 +98,7 @@ class FolderClassifier:
             content = response_dict["choices"][0]["message"]["content"]
             result = FolderClassification.model_validate_json(content)
         except Exception as ex:
-            self.logger.error(f"Failed to parse response: {content}\n{ex}")
+            self.logger.warning(f"Failed to parse response: {content}\n{ex}")
             if '"category": "matter"' in content:
                 result = FolderClassification(category="matter", reasoning="NA")
             else:

{folder_classifier-0.3.4.dist-info → folder_classifier-0.3.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: folder-classifier
-Version: 0.3.4
+Version: 0.3.5
 Summary: Deploy folder classifier API to a Ray cluster
 Author: Crispin Almodovar
 Author-email: 
@@ -11,3 +11,7 @@ Requires-Python: >=3.12
 Description-Content-Type: text/markdown
 
 # Folder classifier API
+
+Implements a REST API for that calls an LLM-based classifier model hosted in Ray serve.
+
+See [Build.md](Build.md) for build and deployment instructions.

folder_classifier-0.3.5.dist-info/RECORD

@@ -0,0 +1,10 @@
+folder_classifier/__init__.py,sha256=k0YWZyUNe7myJiKeX0OaXtJ30_3EGE-vsZiAUbqa-3E,46
+folder_classifier/app.py,sha256=5eaniDWZFBvyI9UN4vOCc6TQwgyptAUVhRVgfrjp2F4,1259
+folder_classifier/classifier.py,sha256=wUoDr8GXajKQPF_zYMeJutpCfuIYVAfAUVjsePfjcIM,8591
+folder_classifier/deploy.py,sha256=06UAxz40IaP28e_RRohJoFwPUrWTaMquGbDylI-oHWA,424
+folder_classifier/dto.py,sha256=KRHEfXKlsIYkBySwHOlo4k5Z4J_F3eR3PeEdUAwj8lI,819
+folder_classifier/util.py,sha256=-Ptxkba5UzmhLrqoiiKZS3G56_cuAMkWlTyHcqdJkg0,3160
+folder_classifier-0.3.5.dist-info/METADATA,sha256=uFoRyzCLh4Gxzak0XktEohvbfHbM7g4Ow3RMoHcZxUs,546
+folder_classifier-0.3.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+folder_classifier-0.3.5.dist-info/top_level.txt,sha256=36ugc9pEbNQ-mnzz4Ot2WVjY3t_LzAN6XOCjDFP4p4k,18
+folder_classifier-0.3.5.dist-info/RECORD,,

folder_classifier-0.3.4.dist-info/RECORD

@@ -1,10 +0,0 @@
-folder_classifier/__init__.py,sha256=k0YWZyUNe7myJiKeX0OaXtJ30_3EGE-vsZiAUbqa-3E,46
-folder_classifier/app.py,sha256=_BUFSR7TJMoylrEwImDlln98C4BciPXoPqM6SQYkc1g,1255
-folder_classifier/classifier.py,sha256=XTqCfUAc4-vX1bS4LixPFNMgvNNFX2mLdWXekHHB-fQ,8285
-folder_classifier/deploy.py,sha256=06UAxz40IaP28e_RRohJoFwPUrWTaMquGbDylI-oHWA,424
-folder_classifier/dto.py,sha256=KRHEfXKlsIYkBySwHOlo4k5Z4J_F3eR3PeEdUAwj8lI,819
-folder_classifier/util.py,sha256=-Ptxkba5UzmhLrqoiiKZS3G56_cuAMkWlTyHcqdJkg0,3160
-folder_classifier-0.3.4.dist-info/METADATA,sha256=YwlhWYE_vMFYfBb0lib6kHUGZjUyH3soNKQiJq0jVmY,392
-folder_classifier-0.3.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-folder_classifier-0.3.4.dist-info/top_level.txt,sha256=36ugc9pEbNQ-mnzz4Ot2WVjY3t_LzAN6XOCjDFP4p4k,18
-folder_classifier-0.3.4.dist-info/RECORD,,