Docling

Docling is IBM’s open-source document-parsing toolkit — it ingests PDFs, Office documents, presentations, images, and scans, and emits a structured representation clean enough to feed into a RAG pipeline, a knowledge base, or an agent’s context window. It’s one of IBM Research’s more quietly important open-source releases; it shows up behind a lot of AI features you might not guess are powered by it.

My work has been on making Docling a first-class citizen inside Langflow and the broader Astra DB ingestion path.

What I’ve shipped

• PR #9398 — Langflow — the initial integration that added Docling as an option in the Langflow File Component, so users can point Langflow at any supported document and get back structured, chunk-ready output without wiring up their own parser.
• PR #12296 — Langflow — fixes to the Docling worker subprocess handling. Running heavy parsers as subprocesses is the kind of thing that works fine in dev and then falls over under real load; this PR tightened the lifecycle and error propagation so Langflow users don’t get silent failures on big files.
• PR #12442 — Langflow — removed the old Astra Assistants path in favour of using Docling directly, simplifying the dependency graph and making document parsing consistent across Langflow surfaces.

Why it matters

Most real-world AI applications spend an embarrassing fraction of their engineering time on the “get the documents into something useful” problem. Docling shrinks that problem to a library call. The integration work is unglamorous glue that, done right, lets everyone above it ship faster — which is exactly the DOE (Developer & Operator Experience) team’s job.