RCOS Project — Spring 2026
Docling-Java is a project aimed at building a local Java API for Docling, a document-processing framework used in AI pipelines. Docling converts unstructured documents into structured, usable data.
Currently, Docling exists only as a Python library. Other languages must access it indirectly through tools like Docling-serve, which exposes it as a REST API over HTTP and runs in Docker.
This approach works but introduces extra overhead and complexity.
The objective of this project is to create a local GraalVM-based Java implementation that allows Java applications to directly interact with the Docling Python library within the same process.
By using GraalVM’s polyglot support, the system will:
- Remove HTTP communication overhead
- Simplify deployment
- Improve performance
- Provide a cleaner developer experience
Existing methods of accessing Docling from Java rely on external communication layers. These add latency, increase system complexity, and make integration harder.
Docling-Java aims to provide:
- Direct integration
- Faster execution
- More maintainable architecture
The project explores GraalVM’s ability to run multiple languages in a single process. This allows Java code to call Python libraries directly without network calls.
Key technical focus areas:
- GraalVM polyglot interoperability
- Embedding Python runtime inside Java
- API design for Java developers
- Performance benchmarking