Generative Computing: An Evening with IBM Research

Explore generative computing with IBM Research in Dublin on 24 September 2026, with technical talks, live coding, open-source AI and real-world Mellea examples. Hosted at Trinity College Dublin Business School from 5:30–9:00pm IST, this Open Source Science Dublin meetup examines how developers can make model calls as structured, testable and governable as conventional function calls. IBM Research hosts the evening, with PyData Ireland participating as a community partner. ## Why attend? Many LLM applications rely on ad-hoc prompts embedded in application code, making systems brittle, difficult to audit and challenging to test. This event introduces Generative Computing as an alternative: treating model calls as first-class program elements that can be typed, composed, verified and governed. Through technical deep dives, demonstrations and cross-disciplinary discussion, the speakers will show how these ideas work in practice: - Generative Computing with Mellea: Programs That Call Models Like Functions , Rahul Nair of IBM Research introduces Mellea, an open-source Python library for predictable generative programs. The session covers instructions, requirements, sampling strategies and verifiers, supported by live coding, practical examples and integration with the wider Python ecosystem. - FactReasoner: Porting a Probabilistic Factuality Assessor to Mellea , Radu Marinescu of IBM Research explains how FactReasoner decomposes long-form generated text into atomic claims and assesses their support against retrieved evidence. The talk explores its move from a bespoke implementation to Mellea, including resulting reliability improvements and speedups. Attendees can also read the associated research paper. - The Mellea Skills Compiler: From Plain-Language Specs to Governed Pipelines , an IBM Research speaker will demonstrate the Mellea Skills Compiler, which turns plain-language skill specifications into governed generative pipelines with evaluation harnesses, requirement checks and provenance. A worked example will explore certification for LLM-backed skills and the path from rapid experimentation to auditable production deployment.… Register for this event