Putting an {ellmer} AI in production with the blessing of IT
summary
Built a working proof of concept in 20 lines of Ellmer code that queries DuckDB and joins Salesforce + Zendesk records into a Shiny app, and users adopt it immediately; InfoSec raises three concrete risks: data leakage of PII and trade secrets, LLM hallucinations that could write false records into production, and model training leakage. Mitigations include AWS Bedrock hosting to prevent retraining on sensitive prompts, strict least‑privilege authorization and scoped write permissions, plus a small browser extension to avoid insecure copy/paste between systems.
AI-written summary · about
related talks
25mSafeguarding Sensitive Data in the Age of Generative AI
06mData Privacy Concerns and Solutions When using LLMs in Production
39mData Privacy Engineering to power Responsible Enterprise AI
53mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
30mBuilding Robust, Secure LLM and Agentic AI Workflows
40mSecuring Generative AI: Is it all an Illusion?
57mAI Applications Security Puzzle
01h 00mWorkshop: Getting Data Ready for AI - With Unified GraphQL &SQL Endpoints