Iris: A Conversational Agent for Complex Tasks
summary
Iris is a conversational data‑science tool that turns Python functions into composable automata via a domain-specific language and a lightweight type system, builds an AST of the session, and exports reproducible Python scripts. Users construct a logistic regression on the Iris dataset, run 10‑fold cross‑validation, and inspect plots via mixed-modality conversation. In an 8‑participant study people complete the modeling task 2.6× faster than with a prepopulated Jupyter Notebook, showing the approach works while highlighting vocabulary and transparency challenges.
AI-written summary · about
chapters (8)
- 0:00Introducing Iris
- 1:48Mixed modality conversation
- 4:37Demo of Iris tool
- 9:10How Iris works
- 11:54User study results
- 13:39Advantages and challenges
- 15:45Future directions
- 16:49Q&A session
related talks
59mMaking the most of artificial and human intelligence for data science
36mGenerative AI + Python: Unlocking Efficiency, Personalization, and Insight
38mSponsored Session: Jupyter AI v3: Human + AI Collaboration
39mJupyter AI — Bringing Generative AI to Jupyter
40mAI Agents in Practice
23mCognitive Architecture for Process industries
26mWelcome to the AI Jungle! Now What?
31mGenerative Conversational AI Interoperability