Proposal: agentic, governance-first evolution of the Information Assistant template (AIA) #994
MarcoPolo483
started this conversation in
Ideas
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hi all,
I've spent the last few months building a proposed next-generation rewrite of the Information Assistant template, published as a working repository for review:
https://github.com/eva-foundry/PubSec-Agentic-InfoAsst
It's called AIA (Agentic Information Assistant) and is framed as a refactor of IA that implements the concepts of The Agentic State (Ilves, Kilian, Parazzoli, Peixoto, Velsberg — Tallinn Digital Summit, 2025). It keeps the IA core — "chat with your own data" over Azure AI Search + Azure OpenAI — and rebuilds it around:
A ReAct-style agentic planner with a governed tool registry (every tool carries a classification ceiling, residency, language support, and HITL flag).
Responsible-AI gates (advisory vs decision-informing outputs; mandatory HITL on decision-informing).
Explainable responses: retrieval path, reasoning trace, negative evidence, source freshness, behavioural fingerprint on every answer.
Multi-language (EN/FR supported) operation with translation as a first-class tool.
FinOps cost attribution wired through APIM with per-workspace cost centres.
A source-authority classifier + conflict resolver for multi-source disagreements.
The README has a table of 13 screenshots covering the full user journey (landing → chat with citations → admin → FinOps/AIOps/LiveOps/DevOps/compliance/red-team), captured via Playwright. Agentic-State-Vision.md maps the architecture to the paper's 12-layer model. AIA-Design-Principles-Beyond-Agentic-State.md captures the architectural gaps the paper leaves open.
What I'm asking: is the maintainer team interested in treating this as a proposal for an IA v2, or for a parallel "agentic" branch of the template? I don't expect a 500-file PR to be reviewable as-is — if there's interest, I'd break it into a sequence of scoped PRs (e.g., provenance/citation layer first, then guardrails, then the planner) with an RFC-style issue up front.
If this is out of scope for the template's direction, a "not for us, but good luck" is a valid response and appreciated.
Thanks,
Marco
Beta Was this translation helpful? Give feedback.
All reactions