2026-08-18-nged-demo
Concerns:
- Context / UX: whether the assistant should automatically pick up dataset context, and how it should behave when users browse datasets without first using chat.
- Cross-dataset joins: how reliably it joins tables, especially with ambiguous or similarly named columns.
- Guardrails / drift: risk of off-topic responses, prompt drift as context grows, and how much monitoring/admin overhead is required.
- Production readiness: whether it’s already live with customers, how close it is to production-ready, and what support/risk-sharing Datopian would provide.
- Security / segregation: customer isolation, avoiding cross-portal data leaks, and support for authenticated/private data.
- Scalability / cost: concurrent usage limits, hosting model, and who carries token/cost overrun risk.
- Integration effort: how much work is needed to embed it into National Grid’s Angular frontend and whether documentation exists.
- Authentication / persistence: whether it can integrate with Amazon Cognito and support persistent chat history across sessions.
- Policy / governance: how Datopian would keep the agent aligned with National Grid’s evolving AI policies and guardrails.
- Advanced capabilities: whether it can handle private user uploads, geospatial/postcode use cases, and bulk metadata handling for existing datasets.
