polis-recognizer 0.1.0__tar.gz

polis_recognizer-0.1.0/.gitignore

@@ -0,0 +1,38 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+dist/
+*.egg-info/
+*.egg
+MANIFEST
+
+# Virtual envs
+venv/
+.venv/
+env/
+
+# Testing
+.pytest_cache/
+.coverage
+htmlcov/
+.ruff_cache/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+.DS_Store
+
+# OS
+Thumbs.db
+
+# Test fixtures generated at runtime
+tests/fixtures/_generated/
+
+# Local-only PDFs (don't commit any real policy)
+*.pdf
+!tests/fixtures/**/*.pdf

polis_recognizer-0.1.0/CHANGELOG.md

@@ -0,0 +1,44 @@
+# Changelog
+
+All notable changes to this project are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.1.0] — 2026-05-03
+
+Initial public release.
+
+### Added
+
+- `PolicyExtractor` facade for end-to-end PDF → 7 structured fields.
+- `extract_from_pdf` / `extract_from_bytes` / `extract_from_text` entry points.
+- 7 deterministic field parsers:
+  - `policy_period`
+  - `franchise`
+  - `limit`
+  - `repair_mode`
+  - `premium`
+  - `sum_type`
+  - `policy_number`
+- Three PDF extractor backends: `pypdf`, `pdfplumber`, `hybrid` (default).
+  - Hybrid mode reuses pypdf text and pdfplumber tables in one pass for the
+    best recall/latency trade-off on KASKO templates.
+- Tesseract OCR fallback for scanned PDFs.
+- OpenCV-based image preprocessing (`fallback` / `always` / `never` modes).
+- Pre-built parser patterns for major Russian KASKO insurers:
+  АльфаСтрахование XLS forms (5/3/5/2 and 5/3/7/2 numbers, branch-letter
+  variants), СОГАЗ-АВТО (`SGZA…` policy numbers), Чулпан (OCR pipe
+  tolerance in policy_number), Ингосстрах (legacy `RUR` currency code,
+  prose-spaced premium label), ВСК (two-row КАСКО layout),
+  АбсолютСтрахование (glued text-layer detection), Diadoc/Kontur EDI
+  envelope detection.
+- Lower-level `run_extraction(text, *, tables=None)` for use without a PDF.
+
+### Notes
+
+- API is pre-stable. Public dataclass shapes
+  (`ExtractedPolicy`, `MonetaryField`, etc.) may change before 1.0.
+- KASKO-only for now; ОСАГО support is on the roadmap.
+
+[0.1.0]: https://github.com/grigra27/polis-recognizer/releases/tag/v0.1.0

polis_recognizer-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,68 @@
+# Contributing
+
+Thanks for your interest in `polis-recognizer`. The project is
+maintained as a side-project, so reviews aren't instant — but PRs are
+welcome.
+
+## What contributions are most useful
+
+The single most valuable contribution is a **parser pattern for an
+insurer format we don't recognize yet**. The recognizer ships with
+patterns for the major Russian KASKO insurers (АльфаСтрахование, СОГАЗ,
+Чулпан, Ингосстрах, ВСК, АбсолютСтрахование, Росгосстрах). Other
+insurers, or unusual templates from these, often don't extract well.
+
+## How to report a new format
+
+**Don't paste real policy PDFs in the issue tracker** — they're personal
+data. Instead:
+
+1. Open an issue titled `New format: <Insurer name>`.
+2. Paste the **plain-text** extraction of the relevant rows
+   (you can run it yourself with
+   `from polis_recognizer import PolicyExtractor; print(PolicyExtractor().extract_from_pdf("polis.pdf"))`
+   and copy the relevant text from the diagnostics, OR run
+   `pdftotext` on the file).
+3. List which fields are missing or wrong, with the expected values.
+4. Anonymize anything that identifies a real client (insured name,
+   address, VIN, policy number — replace with placeholders).
+
+That's enough to write a regex/table pattern. We don't need the
+original PDF.
+
+## How to add a parser pattern
+
+Each field has its own parser at
+`polis_recognizer/extraction/parsers/<field>.py`. The structure is a
+list of `(pattern_id, regex, pattern_strength, context_strength)`
+tuples. Add yours, run the test suite, and open a PR.
+
+Patterns should be **specific** — a pattern that matches "anything that
+looks like a policy number" will produce false positives across the
+corpus. Anchor on insurer-specific labels or layout artifacts when
+possible.
+
+## Testing
+
+```bash
+# Install dev dependencies
+pip install -e ".[test]"
+
+# Run tests
+pytest
+```
+
+The test suite uses synthetic PDFs generated via `reportlab` — no real
+policies are committed to the repo.
+
+## Code style
+
+```bash
+ruff check polis_recognizer tests
+ruff format polis_recognizer tests
+```
+
+## License
+
+By contributing, you agree your contribution is licensed under the
+project's [MIT License](LICENSE).

polis_recognizer-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Grigorii Grachev
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

polis_recognizer-0.1.0/PKG-INFO

@@ -0,0 +1,192 @@
+Metadata-Version: 2.4
+Name: polis-recognizer
+Version: 0.1.0
+Summary: Russian KASKO insurance policy field extractor — pulls 7 structured fields from PDF policies (text-layer + tables + OCR fallback).
+Project-URL: Homepage, https://github.com/grigra27/polis-recognizer
+Project-URL: Repository, https://github.com/grigra27/polis-recognizer
+Project-URL: Issues, https://github.com/grigra27/polis-recognizer/issues
+Project-URL: Changelog, https://github.com/grigra27/polis-recognizer/blob/main/CHANGELOG.md
+Author-email: Grigorii Grachev <grigorii@example.com>
+License: MIT
+License-File: LICENSE
+Keywords: extraction,insurance,kasko,ocr,pdf,policy,russian
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Natural Language :: Russian
+Classifier: Operating System :: MacOS :: MacOS X
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Office/Business
+Classifier: Topic :: Text Processing
+Requires-Python: >=3.11
+Requires-Dist: opencv-python-headless<5,>=4.8
+Requires-Dist: pdf2image<2,>=1.16
+Requires-Dist: pdfplumber<1,>=0.11.0
+Requires-Dist: pillow<12,>=10.0
+Requires-Dist: pypdf<6,>=3.17.4
+Requires-Dist: pytesseract<1,>=0.3.10
+Provides-Extra: dev
+Requires-Dist: pytest-cov>=5.0; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: reportlab>=4.0; extra == 'dev'
+Requires-Dist: ruff>=0.5; extra == 'dev'
+Provides-Extra: test
+Requires-Dist: pytest-cov>=5.0; extra == 'test'
+Requires-Dist: pytest>=8.0; extra == 'test'
+Requires-Dist: reportlab>=4.0; extra == 'test'
+Description-Content-Type: text/markdown
+
+# polis-recognizer
+
+A deterministic field extractor for Russian KASKO insurance policy PDFs.
+Pulls 7 structured fields without LLMs — text-layer extraction
+(`pypdf`) plus optional table-aware reading (`pdfplumber`), with
+Tesseract OCR fallback for scanned policies.
+
+> **Status: pre-stable (0.x).** API may change before 1.0.
+
+## What it extracts
+
+| Field | Type | Example |
+|---|---|---|
+| `policy_number` | `str` | `"AC524160804"` |
+| `policy_period` | `{start, end}` (`date`) | `{"start": date(2025, 2, 27), "end": date(2026, 2, 26)}` |
+| `franchise` | `{value, currency, absent}` | `{"value": 30000, "currency": "RUB", "absent": False}` |
+| `limit` | `{value, currency}` | `{"value": 5525000, "currency": "RUB"}` |
+| `premium` | `{value, currency}` | `{"value": 220000, "currency": "RUB"}` |
+| `sum_type` | `"aggregate"` / `"non_aggregate"` | `"non_aggregate"` |
+| `repair_mode` | `"dealer"` / `"service"` / `"cash"` | `"dealer"` |
+
+## Quick start
+
+```python
+from polis_recognizer import PolicyExtractor
+
+extractor = PolicyExtractor()
+result = extractor.extract_from_pdf("/path/to/polis.pdf")
+
+print(result.policy_number)
+# → "AC524160804"
+print(result.policy_period)
+# → {"start": date(2025, 2, 27), "end": date(2026, 2, 26)}
+print(result.franchise)
+# → {"value": 30000.0, "currency": "RUB", "absent": False}
+```
+
+Input methods:
+
+```python
+extractor.extract_from_pdf("polis.pdf")
+extractor.extract_from_bytes(pdf_bytes, filename="polis.pdf")
+extractor.extract_from_text("сырой текст полиса")  # bypass PDF/OCR
+```
+
+## Installation
+
+```bash
+pip install polis-recognizer
+```
+
+### System dependencies
+
+`polis-recognizer` shells out to Tesseract for OCR and to `poppler`
+for PDF→image conversion. These are NOT pip-installable; install them
+through your OS package manager.
+
+**Linux (Debian/Ubuntu):**
+
+```bash
+sudo apt-get install -y tesseract-ocr tesseract-ocr-rus poppler-utils libgl1
+```
+
+**macOS (Homebrew):**
+
+```bash
+brew install tesseract tesseract-lang poppler
+```
+
+**Windows:** best-effort. Install
+[Tesseract for Windows](https://github.com/UB-Mannheim/tesseract/wiki)
+and add it to PATH; install Poppler binaries for PDF support. We don't
+test on Windows in CI.
+
+The Russian language pack (`tesseract-ocr-rus` / `tesseract-lang`) is
+required — without it, OCR silently falls back to English and Cyrillic
+documents come back as garbage. The library logs a CRITICAL warning at
+import time if the pack is missing.
+
+## How it works
+
+The extractor runs three stages:
+
+1. **PDF ingestion** — `PdfExtractionRouter` tries text-layer extraction
+   first (pypdf for text, pdfplumber for tables on the same page). If
+   the result is too short (fewer than 100 chars by default) or detected
+   as glued/EDI-envelope text, it falls back to Tesseract OCR.
+2. **Text normalization** — Unicode NFKC, NBSP stripping, hyphenated
+   line-break healing, runs of multiple spaces collapsed.
+3. **Field extraction** — 7 deterministic parsers (one per field) run
+   regex + table-aware patterns and emit ``Candidate``s with confidence
+   scores. A ranker picks the winner per field.
+
+There's no LLM and no cloud dependency. Everything runs locally.
+
+## PDF extractor choice
+
+The default is ``"hybrid"`` — pypdf text plus pdfplumber tables in one
+pass. Two alternatives:
+
+| Option | When to use |
+|---|---|
+| `"hybrid"` (default) | Best for KASKO. pypdf preserves date/period text quality, pdfplumber's tables fix the column layout for limit/franchise/premium. |
+| `"pypdf"` | Faster, no tables. Use when document quality is uniform and tables aren't needed. |
+| `"pdfplumber"` | Fully layout-aware. Slower; on KASKO it slightly regresses date parsing. Use for table-heavy non-KASKO formats. |
+
+```python
+extractor = PolicyExtractor(pdf_extractor="pypdf")
+```
+
+## Supported insurer formats
+
+The parser ships with patterns for these Russian insurers' KASKO
+templates: АльфаСтрахование (XLS form-mask), СОГАЗ-АВТО, Чулпан,
+Ингосстрах, ВСК, АбсолютСтрахование, Росгосстрах, СОГАЗ Diadoc-wrapped
+PDFs. Recall on real-world KASKO corpora is ~50-65% per field; pulling
+above that requires per-format parser additions.
+
+If you have a policy from an insurer not on this list — see
+[CONTRIBUTING.md](CONTRIBUTING.md) for how to add a parser pattern.
+
+## Configuration
+
+Constructor arguments:
+
+```python
+extractor = PolicyExtractor(
+    ocr_language="rus+eng",       # Tesseract language string
+    ocr_timeout_seconds=300,
+    ocr_page_limit=50,
+    ocr_max_text_size=500_000,
+    pdf_extractor="hybrid",       # "pypdf" | "pdfplumber" | "hybrid"
+    image_preprocessing="fallback", # "never" | "fallback" | "always"
+    psm=None,                     # Tesseract --psm (None = auto)
+    oem=None,                     # Tesseract --oem (None = auto)
+    max_image_size_bytes=None,    # reject images larger than this
+)
+```
+
+## License
+
+[MIT](LICENSE) © Grigorii Grachev. Free for any use, including
+commercial.
+
+## Roadmap
+
+This 0.1.0 release focuses on KASKO. ОСАГО support is on the roadmap —
+the underlying field model already accommodates it; only parser
+patterns need to be added.
+
+See [CHANGELOG.md](CHANGELOG.md) for release history.

polis_recognizer-0.1.0/README.md

@@ -0,0 +1,151 @@
+# polis-recognizer
+
+A deterministic field extractor for Russian KASKO insurance policy PDFs.
+Pulls 7 structured fields without LLMs — text-layer extraction
+(`pypdf`) plus optional table-aware reading (`pdfplumber`), with
+Tesseract OCR fallback for scanned policies.
+
+> **Status: pre-stable (0.x).** API may change before 1.0.
+
+## What it extracts
+
+| Field | Type | Example |
+|---|---|---|
+| `policy_number` | `str` | `"AC524160804"` |
+| `policy_period` | `{start, end}` (`date`) | `{"start": date(2025, 2, 27), "end": date(2026, 2, 26)}` |
+| `franchise` | `{value, currency, absent}` | `{"value": 30000, "currency": "RUB", "absent": False}` |
+| `limit` | `{value, currency}` | `{"value": 5525000, "currency": "RUB"}` |
+| `premium` | `{value, currency}` | `{"value": 220000, "currency": "RUB"}` |
+| `sum_type` | `"aggregate"` / `"non_aggregate"` | `"non_aggregate"` |
+| `repair_mode` | `"dealer"` / `"service"` / `"cash"` | `"dealer"` |
+
+## Quick start
+
+```python
+from polis_recognizer import PolicyExtractor
+
+extractor = PolicyExtractor()
+result = extractor.extract_from_pdf("/path/to/polis.pdf")
+
+print(result.policy_number)
+# → "AC524160804"
+print(result.policy_period)
+# → {"start": date(2025, 2, 27), "end": date(2026, 2, 26)}
+print(result.franchise)
+# → {"value": 30000.0, "currency": "RUB", "absent": False}
+```
+
+Input methods:
+
+```python
+extractor.extract_from_pdf("polis.pdf")
+extractor.extract_from_bytes(pdf_bytes, filename="polis.pdf")
+extractor.extract_from_text("сырой текст полиса")  # bypass PDF/OCR
+```
+
+## Installation
+
+```bash
+pip install polis-recognizer
+```
+
+### System dependencies
+
+`polis-recognizer` shells out to Tesseract for OCR and to `poppler`
+for PDF→image conversion. These are NOT pip-installable; install them
+through your OS package manager.
+
+**Linux (Debian/Ubuntu):**
+
+```bash
+sudo apt-get install -y tesseract-ocr tesseract-ocr-rus poppler-utils libgl1
+```
+
+**macOS (Homebrew):**
+
+```bash
+brew install tesseract tesseract-lang poppler
+```
+
+**Windows:** best-effort. Install
+[Tesseract for Windows](https://github.com/UB-Mannheim/tesseract/wiki)
+and add it to PATH; install Poppler binaries for PDF support. We don't
+test on Windows in CI.
+
+The Russian language pack (`tesseract-ocr-rus` / `tesseract-lang`) is
+required — without it, OCR silently falls back to English and Cyrillic
+documents come back as garbage. The library logs a CRITICAL warning at
+import time if the pack is missing.
+
+## How it works
+
+The extractor runs three stages:
+
+1. **PDF ingestion** — `PdfExtractionRouter` tries text-layer extraction
+   first (pypdf for text, pdfplumber for tables on the same page). If
+   the result is too short (fewer than 100 chars by default) or detected
+   as glued/EDI-envelope text, it falls back to Tesseract OCR.
+2. **Text normalization** — Unicode NFKC, NBSP stripping, hyphenated
+   line-break healing, runs of multiple spaces collapsed.
+3. **Field extraction** — 7 deterministic parsers (one per field) run
+   regex + table-aware patterns and emit ``Candidate``s with confidence
+   scores. A ranker picks the winner per field.
+
+There's no LLM and no cloud dependency. Everything runs locally.
+
+## PDF extractor choice
+
+The default is ``"hybrid"`` — pypdf text plus pdfplumber tables in one
+pass. Two alternatives:
+
+| Option | When to use |
+|---|---|
+| `"hybrid"` (default) | Best for KASKO. pypdf preserves date/period text quality, pdfplumber's tables fix the column layout for limit/franchise/premium. |
+| `"pypdf"` | Faster, no tables. Use when document quality is uniform and tables aren't needed. |
+| `"pdfplumber"` | Fully layout-aware. Slower; on KASKO it slightly regresses date parsing. Use for table-heavy non-KASKO formats. |
+
+```python
+extractor = PolicyExtractor(pdf_extractor="pypdf")
+```
+
+## Supported insurer formats
+
+The parser ships with patterns for these Russian insurers' KASKO
+templates: АльфаСтрахование (XLS form-mask), СОГАЗ-АВТО, Чулпан,
+Ингосстрах, ВСК, АбсолютСтрахование, Росгосстрах, СОГАЗ Diadoc-wrapped
+PDFs. Recall on real-world KASKO corpora is ~50-65% per field; pulling
+above that requires per-format parser additions.
+
+If you have a policy from an insurer not on this list — see
+[CONTRIBUTING.md](CONTRIBUTING.md) for how to add a parser pattern.
+
+## Configuration
+
+Constructor arguments:
+
+```python
+extractor = PolicyExtractor(
+    ocr_language="rus+eng",       # Tesseract language string
+    ocr_timeout_seconds=300,
+    ocr_page_limit=50,
+    ocr_max_text_size=500_000,
+    pdf_extractor="hybrid",       # "pypdf" | "pdfplumber" | "hybrid"
+    image_preprocessing="fallback", # "never" | "fallback" | "always"
+    psm=None,                     # Tesseract --psm (None = auto)
+    oem=None,                     # Tesseract --oem (None = auto)
+    max_image_size_bytes=None,    # reject images larger than this
+)
+```
+
+## License
+
+[MIT](LICENSE) © Grigorii Grachev. Free for any use, including
+commercial.
+
+## Roadmap
+
+This 0.1.0 release focuses on KASKO. ОСАГО support is on the roadmap —
+the underlying field model already accommodates it; only parser
+patterns need to be added.
+
+See [CHANGELOG.md](CHANGELOG.md) for release history.

polis_recognizer-0.1.0/examples/basic_usage.py

@@ -0,0 +1,43 @@
+"""Minimal example — extract structured fields from a single policy PDF."""
+
+from polis_recognizer import PolicyExtractor
+
+
+def main(pdf_path: str) -> None:
+    extractor = PolicyExtractor()
+    result = extractor.extract_from_pdf(pdf_path)
+
+    print(f"Policy number:     {result.policy_number or '—'}")
+    if result.policy_period:
+        print(
+            f"Policy period:     {result.policy_period['start']} → "
+            f"{result.policy_period['end']}"
+        )
+    if result.franchise:
+        if result.franchise.get("absent"):
+            print(f"Franchise:         absent (no deductible)")
+        else:
+            print(
+                f"Franchise:         {result.franchise['value']} "
+                f"{result.franchise['currency']}"
+            )
+    if result.limit:
+        print(f"Sum insured:       {result.limit['value']} {result.limit['currency']}")
+    if result.premium:
+        print(
+            f"Premium:           {result.premium['value']} {result.premium['currency']}"
+        )
+    print(f"Sum type:          {result.sum_type or '—'}")
+    print(f"Repair mode:       {result.repair_mode or '—'}")
+    print()
+    print(f"Extraction method: {result.extraction_method}")
+    print(f"Status:            {result.extraction_status}")
+
+
+if __name__ == "__main__":
+    import sys
+
+    if len(sys.argv) != 2:
+        print("Usage: python examples/basic_usage.py /path/to/policy.pdf")
+        sys.exit(1)
+    main(sys.argv[1])

polis_recognizer-0.1.0/polis_recognizer/__init__.py

@@ -0,0 +1,98 @@
+"""polis-recognizer — Russian insurance policy field extractor.
+
+Extracts 7 structured fields from KASKO/insurance policy PDFs:
+``policy_period``, ``franchise``, ``limit``, ``repair_mode``,
+``premium``, ``sum_type``, ``policy_number``.
+
+Quick start::
+
+    from polis_recognizer import PolicyExtractor
+
+    extractor = PolicyExtractor()
+    result = extractor.extract_from_pdf("/path/to/polis.pdf")
+
+    print(result.policy_number)
+    print(result.policy_period.start, result.policy_period.end)
+    print(result.franchise.value, result.franchise.currency)
+
+The extractor combines a text-layer reader (pypdf) with optional
+table-aware extraction (pdfplumber) and an OCR fallback (Tesseract).
+See README.md for the full API and supported insurer formats.
+"""
+
+from .contract_field_extractor import (
+    ContractFieldExtractor,
+    ContractFieldsResult,
+    FieldDiagnostic,
+    MonetaryField,
+    PolicyPeriodField,
+    TextField,
+)
+from .exceptions import (
+    OCRProcessingError,
+    OCRTimeoutError,
+    UnsupportedFileTypeError,
+)
+from .extraction import (
+    Candidate,
+    ExtractionV2Result,
+    run_extraction,
+)
+from .extractor import ExtractedPolicy, PolicyExtractor
+from .hybrid_ingestion import HybridIngestionService
+from .ocr_config import (
+    OCRConfig,
+    OCRResult,
+    get_ocr_config,
+    reset_ocr_config,
+    validate_language_pack,
+)
+from .ocr_service import OCRService
+from .pdf_extraction_router import (
+    PdfExtractionOutcome,
+    PdfExtractionRouter,
+    build_text_service,
+)
+from .pdfplumber_ingestion import PdfPlumberIngestionService
+from .policy_ingestion import ExtractedTextResult, PolicyIngestionService
+
+
+__version__ = "0.1.0"
+
+
+__all__ = [
+    "__version__",
+    # Top-level facade (recommended entry point)
+    "PolicyExtractor",
+    "ExtractedPolicy",
+    # Field result dataclasses
+    "ContractFieldsResult",
+    "FieldDiagnostic",
+    "MonetaryField",
+    "PolicyPeriodField",
+    "TextField",
+    # Lower-level pipeline (for advanced use)
+    "ContractFieldExtractor",
+    "Candidate",
+    "ExtractionV2Result",
+    "run_extraction",
+    # OCR service
+    "OCRService",
+    "OCRConfig",
+    "OCRResult",
+    "get_ocr_config",
+    "reset_ocr_config",
+    "validate_language_pack",
+    # PDF ingestion
+    "PdfExtractionRouter",
+    "PdfExtractionOutcome",
+    "ExtractedTextResult",
+    "PolicyIngestionService",
+    "PdfPlumberIngestionService",
+    "HybridIngestionService",
+    "build_text_service",
+    # Exceptions
+    "OCRProcessingError",
+    "OCRTimeoutError",
+    "UnsupportedFileTypeError",
+]