AI at Anaconda with Greg Jennings

What It Covers

Greg Jennings, VP of Engineering and AI at Anaconda, discusses how the company integrates AI assistants into Jupyter notebooks and Excel workflows, enabling context-aware code generation, error fixing, and democratizing data science capabilities.

Key Questions Answered

• •Context-Aware Notebook Assistant: Anaconda Assistant eliminates copy-paste workflows by automatically injecting notebook context (data frames, columns, variables, errors) into AI prompts, enabling inline code generation and automatic error fixing with over 50,000 active users.
• •Prompt Optimization Over Fine-Tuning: The team achieves reliable code generation through extensive prompt engineering and context injection rather than model fine-tuning, tracking error rates and iteratively adjusting prompts to reduce failures in specific workflows like data visualization.
• •Package Management Evolution: Future package management must expand beyond Python dependencies to include AI models as dependencies, allowing developers to conda install applications with embedded models, agents, and runtime environments through tools like AI Navigator.
• •Evaluation Framework Requirements: Organizations building LLM applications need custom evaluation frameworks based on real user interactions and topic modeling, as even the best models fail unpredictably and require domain-specific testing beyond generic benchmarks to ensure reliability.