Building Governed AI Applications on Palantir Foundry: Turning a Meeting Into an Ontology Object

Every team knows this relay: a meeting happens, the transcript lands somewhere, someone has to go find it, and then the real work starts — deciding what mattered, rebuilding the slides, reapplying the template, reworking one section without breaking the other fourteen. None of that is hard. It's just slow, and it happens the same way every time.

Our team decided to stop treating that relay as unavoidable. The result is Phanes, a meeting-to-deck application we built entirely on Palantir Foundry. The interesting part isn't that it generates slides — it's that we used Foundry to make the deck itself something the platform understands, tracks and governs, rather than a file that happens to have been produced with AI.

SECTION 01
The problem we were actually solving

Meetings live wherever they were held — a drive, a calendar entry, a chat thread. Finding the right one is a search across tools before any work even starts, and whether a transcript exists is usually something you discover by opening the file. Then every deck rebuilds the same narrative shape from scratch. The first draft is the easy part; revising one slide without disturbing the rest is the cost nobody puts on a timeline.

SECTION 02
What we built

01 — Select. Choose a meeting from Google Drive or Google Calendar. Transcript availability is flagged right on the row, before you commit.

02 — Configure. Set the title, file name, instructions, template and model up front. Everything that shapes the output is decided before generation, not discovered after.

03 — Generate. A Foundry Action carries the meeting context and configuration into the generation workflow; the deck returns as an interactive presentation inside the app.

04 — Refine & export. Edit one slide or the whole deck in natural language, then export to HTML or PDF when it's ready to share

SECTION 03
Why Foundry is the interesting part

We could have bolted a slide generator onto a standalone front end. We didn't, because the value we wanted was a deck that stays connected to where it came from, with an edit history nobody has to remember to log. The descriptions below are of Aavya's configuration of Palantir Foundry, not of Foundry's capabilities in the abstract.

THE SAME APPLICATION, SEEN IN FOUNDRY

What “Ontology, not files” looks like as data lineage

Foundry's data lineage view of Phanes, resolved to 33 nodes. Governed source connections and network egress policies feed transcript data on the left; three object types — Transcription, Deck Model Usage and Deck — sit at the centre; nine action types write to them. Nothing edits a document directly — the ontology claim above, rendered as structure rather than assertion. The OSDK workspace on the right reads the same ontology, and five automations hang off the Deck object for sharing, Slack delivery and PDF generation.

Returned as an interactive deck, not handed over as a download

SECTION 04
What this unlocks

Templates as objects — a brand change could propagate to future decks without touching the application.

More sources, same object — adding a new notetaker or chat export becomes an ingestion job, not an application change.

Decks that link onward — linking a deck forward to a project or decision record is what makes it findable months later instead of just filed.

Built by Aavya on Palantir Foundry. This post is published by Aavya and describes Aavya's own product and Aavya's own configuration of Palantir Foundry. It is not published on behalf of Palantir Technologies Inc., does not represent Palantir's views, and is not a statement about Palantir's products or compliance certifications generally. Screenshots are taken from Aavya's own internal tenant and show Aavya's own internal meeting material; personal identifiers and third-party model names are blurred in every image. No customer or production data appears in any image. Statements about identity scoping, source connectivity, recording and sharing describe how Aavya has configured Phanes at the time of writing; they are not security, compliance, certification or capability claims about Palantir Foundry. Press, analyst and investor enquiries relating to Palantir are referred to Palantir. Palantir, Palantir Foundry and Foundry are trademarks of Palantir Technologies Inc.