WOMBAT
  • Schedule
  • Contact
  • About
    • Code of Conduct
    • Organising Committee
  • Register

From Idea to Production: Building an Agentic Forecasting Ecosystem in the Open

Generated summary
Published

December 1, 2026

December 01, 03:30 PM

From Idea to Production: Building an Agentic Forecasting Ecosystem in the Open

Over the past two decades, open-source forecasting software has transformed how we build forecasting workflows. Rather than focusing solely on algorithms, practitioners and researchers have built an ecosystem of open-source forecasting tools, each exposing different interfaces and often even different programming languages. Agentic forecasting builds on this foundation by introducing large language models as orchestration layers that can reason about forecasting tasks while leveraging the rich ecosystem of existing methods. In this talk, we will share the lessons we learned while building TimeCopilot, an open-source agentic forecasting framework developed in the open. We'll explore the engineering decisions behind bringing together foundation models and classical forecasting methods under a common interface, the challenges of making LLM-driven workflows reliable, and the trade-offs involved in designing software that remains transparent and reproducible as the ecosystem evolves. Beyond the architecture itself, we'll discuss what it means to build an AI ecosystem in the open. How do you design APIs for a landscape where models change every few months? How do you balance rapid experimentation with long-term maintainability? And how do you create infrastructure that supports both research prototypes and production deployments? We'll close by reflecting on where agentic forecasting is headed and why the next advances may come as much from software engineering and open collaboration as from the models themselves.
Register for the workshop day
Mitchell O’Hara-Wild

Renée Rosillo

Technologist and entrepreneur working at the intersection of Agentic AI and Machine Learning. Co-Founder of TimeCopilot. Renée also co-founded AO Labs, working on continuous learning weightless neural networks. Published researcher and participant of Creative Destruction Lab Toronto AI at University of Toronto. She previously co-founded Prism, a global network for queer founders, and co-founded Minerva Robotics and FounderFamilia.

Mitchell O’Hara-Wild

Azul Garza

🇲🇽🏳️‍⚧️ AI and Machine Learning leader with deep experience in econometrics and mathematical modeling. Co-Founder of TimeCopilot and co-creator of the Nixtlaverse and TimeGPT. Former CTO and Co-Founder of Nixtla. A frequent live-coding speaker at ISF, Microsoft Build, AWS re:Invent, and PyCons. Published researcher and recipient of multiple forecasting and AI awards.

Workshop Organised by the Monash Business Analytics Team