Public health planning · Reporting & collaboration

One place to read the plan — and understand it at a glance.

AI Overview turns dense Action Plan submissions into a short, structured executive read: grounded in the data teams already entered, easy to scan, and refreshable when the plan changes.

Built into the PHABC Tool and powered by the SundayPyjamas platform API — so assistive synthesis stays on a single managed integration instead of separate tools or blank chat threads.

Public health planning
Reporting & collaboration
SundayPyjamas platform API
In-product synthesis
Canada

PHABC Tool · AI Overview

1 action

Coherent overview on demand

100%

Grounded in submitted plan data

0

Parallel documents to maintain

1 API

Managed platform integration

Design goals for the feature; individual organizations measure impact in their own reporting cycles.

The problem

Action plans are not spreadsheets alone. They combine structured fields with qualitative insight: program descriptions, outcome narratives, indicator wording that does not always reduce to a single metric, and partnership context that shapes how work is interpreted.

Teams were left with two uncomfortable options: skim and risk missing nuance, or invest significant time reconstructing the story by hand before meetings, approvals, or cross-team alignment. The product needed a trustworthy way to process that qualitative layer — alongside structured fields — into something leadership and partners could use between formal reports.

What we built

AI Overview is the in-product answer: synthesis that lives on the Action Plan, not elsewhere. Reviewers can orient quickly, authors can sanity-check whether their intent reads back clearly, and organizations keep one source of truth (the plan) while offering a human-readable layer on top.

What we shipped

  • AI Overview control on the Action Plan — explicit trigger, visible wait, clear success or failure
  • AI Summary panel with readable headings and lists, optional refresh when the plan changes, and a clear empty state
  • Grounded server-side generation: plan content the user may access, formatted as narrative context for the model
  • Session-friendly persistence so returning to the plan does not always require another round trip, with deliberate regenerate when the submission evolves

Technical integration

High-level flow: authenticated request, plan content in context, SundayPyjamas API, summary returned to the page — intentionally simple so future refinements (caching, org libraries, stricter permissions) do not change the core story.

End-to-end flow

Authenticated request → plan in context → SundayPyjamas API → summary

single integration surface — no ad hoc assistants or copy-paste workflows

Security posture

Server-side configuration

sensitive provider configuration never exposed to the browser

Generation contract

Chat-style request + formatted plan context

instructions favor stakeholder-ready headings and bullets — not raw tables or implementation dumps

Product surface

Action Plan page (PHABC Tool)

one button, one overview, tied to this plan — authoritative record stays in-product

Summaries use public-health planning language and structured presentation suitable for stakeholders — not a dump of implementation detail. The pattern keeps the door open for shared caching and org-wide libraries without changing the user story: one button, one overview, tied to this plan.

Outcomes

Program and planning leads: less time reconstructing what the plan actually emphasizes before workshops, steering conversations, or handoffs.

Authors: a quick mirror for how dense qualitative and indicator language might land for someone who did not write every paragraph.

Organizations: a more consistent path from submitted detail to shared understanding — without maintaining a second document in parallel.

What this enables

The same pattern — rich operational data in the product, assistive synthesis beside it, one provider integration — extends to any workflow where qualitative and structured inputs live together and stakeholders still need a credible executive read.

Plans & assessments

Long-form or mixed records with bounded context to a managed AI API and first-class UI on the object.

Portfolios & compliance

Authoritative record stays in your product; overview layers refresh when the underlying submission changes.

Cross-team alignment

Executive reads without asking teams to duplicate narrative in slide decks or external assistants.

Building something similar?

Keep the record authoritative in your product, send bounded context to a managed AI API, and render results as first-class UI so users never leave the object they care about.