From Unstructured Documents to Actionable Insights: A Workshop on Docling, Granite, and Mellea

Free Workshop

9:00 am – 12 noon

Abstract

AI helps us make sense of the world’s knowledge, yet much of that knowledge remains trapped in unstructured formats—PDFs, scanned reports, charts, and complex tables. Even after extraction, deriving true meaning requires more than a simple query to an LLM. This workshop bridges two open-source paradigms to address that gap: first, by liberating documents with Docling and the Granite family of efficient small language models, then by orchestrating that intelligence with Mellea’s generative programming model—again powered by Granite as the composable reasoning engine.

Register

---

Part I: Document Intelligence with Docling and Granite

We begin with a hands-on exploration of Docling, an open-source library that converts raw documents into richly structured data — detecting reading order, tables, formulas, OCR text, and more. Through progressive examples, participants will move from a basic multimodal RAG pipeline (combining text and images for LLM context) to interpreting complex charts using chart2csv, a specialised model trained to translate visual charts into usable datasets. Every generative step in this segment is powered by Granite small language models, chosen for their efficiency, transparency, and strong performance on enterprise document tasks. The session culminates in an advanced agentic flow: Docling as an MCP tool, where an autonomous agent decides when and how to invoke document parsing. By the end of this segment, attendees will know how to build multimodal, agent-ready systems on top of messy, real-world documents—using only open-source components.

---

Part II: Generative Programming with Mellea and Granite

Building on the structured outputs from Docling, the second half introduces Mellea, an open-source framework that decomposes AI behavior into explicit, composable steps. We begin with the philosophy behind generative programming and Mellea’s core architecture, then explore its advanced capabilities: multi-agent coordination, tool integration, and sophisticated workflow automation. Throughout, Granite small language models serve as the generative workhorses—called by Mellea to reason, plan, and generate code or insights on demand. The highlight is a guided, hands-on exercise that unites the entire stack: attendees will use Mellea to orchestrate Granite models and the Docling pipelines built earlier, transforming unstructured information into actionable insights through a clean, governable, and reproducible generative program.

---

Who Should Attend

This workshop is designed for software architects, senior developers, AI/ML engineers, and technical leaders who want to move beyond ad‑hoc LLM wrappers. Come see how the combination of explicit document structuring, efficient open models, and composable generative logic can reshape the way we build intelligent, trustworthy systems—end to end, in the open.