← Mimir docs

Mimir — architecture & product review

What we're thinking, and what we want to decide together.

Format

Leadership review · ~60 min + discussion

In the room

Engineers & Architects · ML team · CTO · Chief Strategy Officer · VP Engineering · PM

Goal

Align on the platform shape and get a go/no-go on Phase 1

Pre-read

The docs hub + the back-test, ideally before the room

The decision we're here to make: do we commit to building Phase 1 — one real product, end to end, on the Mimir engine — with the pack boundary as the explicit success test? Plus: resourcing for Phase 1, and a greenlight for a first design-partner pilot.

Running order

#TopicWhat we coverTime
1Why we're hereThe problem shape, and the decision on the table5 min
2The thesisOne engine, domain knowledge as packs; the two planes; the component map8 min
3One case, end to endThe 8-step example, then "swap the pack" → a different product, same engine8 min
4How we pressure-tested itThe back-test: 13 products on paper → 18 requirements; what it forced us to add8 min
5Execution, models & complianceSync/async + priority; Model Gateway; compliance as a USP (and its real build items)8 min
6Build phases & buy-vs-buildPhase 1→3; what we integrate (commodities) vs what we own (the moat)7 min
7Risks & open questionsThe honest thin-ice list — where we are not yet proven8 min
8The askWhat we need from each role to start Phase 16 min
9Discussion & next stepsDecision, owners, follow-upsopen

What we want your read on

Open questions for the room

  • Do we agree the buy-vs-build line — integrate OCR / vision / LLM / workflow / model-gateway; build the Signal model, fused Detect, pack & compliance layer?
  • Which domain is pack #2 (health · lending · KYC)? It's how we prove "domain-agnostic."
  • Risk appetite for the Agent Runtime and the cross-tenant link graph (both Phase 3)?
  • Compliance priorities — which regimes / certifications matter first?

Not deciding today

  • The final product / market name (separate track).
  • Pricing & packaging.
  • The cross-tenant link-graph design — Phase 3, needs legal review.
  • Any published-accuracy number — nothing is measured yet.
Honesty note for the room: every figure on the slides is illustrative; the 13 products are a design exercise, not shipped software. No named customers, sample data throughout. "Agentic" and the link graph are roadmap, not built. "Domain-agnostic" is design intent until pack #2 ships with zero engine change.

Pre-reads: the docs hub · the back-test (one page) · the architecture map · the deck

Mimir · the Modules intelligence engine — review agenda (draft).