Tools for processing scanned PDFs

Dedoc is an open-source Python system built for developers to parse PDFs, Office files and scanned images with Tesseract OCR plus ML models for layout and structure extraction; it explicitly extracts logical structure and tables and can run as a Docker-hosted REST API, matching your need for OCR, table-aware extraction, varied layouts, and batch integration into a pipeline.

Best for: Teams that want a self-hosted, developer-friendly OCR+layout parser they can run in Docker and integrate into ETL jobs.

Caveat: The project notes it only processes black-and-white scans with its image handler, so color-scan handling may be limited.

Won’t cover: It doesn't provide a built-in Postgres import connector so you'll need to map its outputs to your Postgres ingestion step.

OpenDataLoader PDF is an open-source parser that detects tables, tagged structure and layout and emits clean HTML, Markdown or JSON suitable for machine processing; its focus on preserving table and layout structure makes it a strong fit for extracting tables reliably from scanned/complex PDFs and feeding structured outputs into a database-backed system.

Best for: Developers who need layout-aware extraction that preserves tables and produces JSON/HTML outputs for downstream ingestion.

Won’t cover: It doesn't include a native Postgres importer so you'll need to convert its JSON/HTML output into your Postgres schema.

DocuPipe is positioned as an API-first intelligent document processing platform that performs OCR, extracts tables into rows/columns, and returns clean structured data—matching your need for OCR on scanned PDFs, table extraction, and an API-driven workflow suitable for bulk ingestion.

Best for: Developers who prefer a managed/API-first extractor that returns structured table and field data for pipeline integration.

Won’t cover: It doesn't advertise a built-in Postgres connector so you'll need to map API responses into your Postgres import process.

ICIJ Extract is a CLI designed for parallelized, distributed extraction and OCR across large heterogeneous document sets; it uses a Redis queue to distribute work and was built for processing millions of documents, which aligns with your scale, splitting and batch-processing requirements.

Best for: Engineering teams that need a battle-tested, distributed extraction CLI to run large-scale OCR and text/metadata extraction across many machines.

Caveat: Its documented outputs are plain text, Solr, or stdout, so direct connectors to relational databases are not part of its core outputs.

Won’t cover: It doesn't provide a native Postgres import target, so you must transform its text/metadata outputs into a Postgres-friendly format.

Unstract is built to turn unstructured documents (PDFs, images, scans) into structured JSON using prompt-driven extraction and can be deployed as an API or ETL pipeline; that model-driven approach is useful for generating consistent fields, titles, and structured table-like outputs across heterogeneous layouts.

Best for: Teams that want schema-driven, developer-configurable extraction (via natural-language prompts) and an API/ETL deployment model for integration into Postgres-backed systems.

Won’t cover: It doesn't include a direct Postgres connector so you'll need to map its JSON outputs into your database.

jt-doc-tools is a self-hosted PDF/Office toolbox that bundles OCR, page splitting/merging, and many PDF utilities with a REST API and enterprise features; its broad toolset fits workflows that require splitting multi-document PDFs, OCRing pages, and automating conversions at scale on-premises.

Best for: Teams wanting an all-in-one self-hosted toolkit to run OCR, split/merge PDFs and expose processing via an API for downstream ingestion.

Caveat: The project is an extensive, heavier all-in-one platform and requires significant disk/host resources compared with single-purpose microservices.

Won’t cover: It doesn't offer a turnkey Postgres import, so you'll need to implement the final mapping into your Postgres schema.