In Atlanta, Docling workshops put documents to work for AI
· BehindLines newsroom
Before an assistant can answer questions from a collection of documents, it needs a usable representation of their structure. That is the problem addressed by Docling workshops on the programme of IBM TechXchange, scheduled for 26–29 October 2026 in Atlanta. The calendar assigns Monday 26 to pre-conference activities and lists the opening general session for Tuesday 27. IBM programme
Keeping a page’s meaning
Docling is an open-source project originating at IBM Research and hosted by the LF AI & Data Foundation. The schedules cited here come from the project and IBM. Project background
A table is not read in the same way as a paragraph: moving a value away from its column can change what it means. Docling’s documentation describes a document-processing tool that handles PDFs, including their layout, tables and reading order. Structured information of this kind can then feed an artificial intelligence application. Documentation
The project’s published schedule includes seven sessions: four practical labs and three talks. This is a training programme, rather than an announcement of a new software version at the conference. Docling schedule
From documents to information retrieval
The calendar lists Monday 26’s lab at 12:15–13:45 local time, EDT. It combines Docling with OpenRAG and an OpenSearch knowledge base, which agents can query through the MCP protocol. The stated aim is to build a pipeline connecting documents with the retrieval tools used by assistants. IBM session details
Tuesday 27’s programme includes another lab using Amazon Bedrock and S3 Vectors. A Wednesday 28 session is due to follow a document arriving in storage and its subsequent processing with Kafka and Confluent. These are scheduled exercises; no outcome from those sessions has been established here. Calendar
A starting point for developers
The public materials repository contains directories for labs LAB-1499 and LAB-2775. It provides a concrete way to inspect exercises beyond the conference itself. Materials
The common thread matters: supplying documents to AI also means preserving relationships within them. A successful retrieval demonstration remains an exercise. By itself, it cannot establish that every answer from an application will be reliable.