Add ocr service

[SahilBheke25]

No description provided.

[Copilot]

Pull request overview

Adds a new standalone OCR/PDF text-extraction module under OCR/, designed to be reusable as both a CLI tool and an importable package with a pluggable OCR-engine interface (currently Surya OCR).

Changes:

• Introduces the extractor/ package with a generic OCR result model, load/format pipeline, and engine registry + Surya engine adapter.
• Adds a CLI entry point (OCR/main.py) and extensive module documentation (OCR/README.md).
• Adds module dependencies (OCR/requirements.txt) and updates repo .gitignore.

Reviewed changes

Copilot reviewed 11 out of 14 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
.gitignore	Adds an ignore entry for macOS metadata files (currently with incorrect casing).
OCR/README.md	Documents goals, architecture, usage, output formats, and extension points for the OCR module.
OCR/main.py	Provides a CLI wrapper around Extractor.run() with --engine selection.
OCR/requirements.txt	Declares OCR module dependencies (currently unpinned).
OCR/extractor/__init__.py	Exposes the public API (Extractor, defaults, ExtractionResult).
OCR/extractor/pipeline.py	Implements the high-level extraction pipeline and file output naming.
OCR/extractor/models.py	Defines engine-agnostic Block and PageResult dataclasses.
OCR/extractor/loader.py	Implements PDF discovery and PDF→image rendering via pypdfium2.
OCR/extractor/formatter.py	Builds plain text, JSON, and HTML outputs from engine-agnostic results.
OCR/extractor/engines/__init__.py	Adds the engine registry + get_engine() factory.
OCR/extractor/engines/base.py	Defines the BaseOCREngine interface contract.
OCR/extractor/engines/surya_engine.py	Implements the Surya OCR adapter and lazy backend initialization.

Comments suppressed due to low confidence (2)

OCR/README.md:310

• The README claims the render output is “capped at ~2048px width”, but loader.py currently renders at a fixed scale and does not enforce any max-width cap. Either implement the cap in code or adjust the docs to avoid promising behavior that isn’t present.

Since sample documents are in Hindi/Marathi/mixed scripts:
- Render scale defaults to produce ~150–200 DPI equivalent (capped at ~2048px width) for legible text — this is the single biggest accuracy lever per Surya's own docs.

OCR/extractor/formatter.py:121

• The HTML report injects block.html directly into the output. Since OCR output ultimately comes from untrusted PDFs, this can embed unexpected tags (e.g., <script>), which will execute if the report is opened in a browser or served by a web app. Consider sanitizing to an allowlist (tables/paragraphs/etc.) before writing HTML.

        for block in page_result.blocks:
            if block.skipped:
                continue
            parts.append("<div class='block'>")
            parts.append(block.html)
            parts.append("</div>")

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.