New archetype

Governance Archetype

Portfolio governance workflows — structured intake, human approval gates, research, material generation, and post-approval portfolio refresh — built on AESOP’s existing pipeline engine. No new infrastructure required.

intake → approver gate → research → materials → governance gate → refresh → monitor → review gate → apply → track

01Pipeline flow

1Capture

Intake. Discovery interview plus gap analysis on the submitted project.

Gate — intake

An approver reviews intake completeness.

approvedProceed to research.

rejectedReturned for revision.

2Research

Deep research plus market analysis and a regulatory scan.

3Materials

Generate materials. One-pager, pro forma, and the PPTX governance deck.

Gate — governance

The governance body reviews.

approvedProceed to portfolio refresh.

rejectedMaterials revised.

4Refresh

Portfolio refresh. Update the system of record and regenerate the Excel / PPTX reports.

5Monitor

Meeting monitor. Ingest Teams transcripts and run LLM extraction of status updates.

Gate — review

A human reviewer approves or rejects each proposed update.

approvedApply the change.

rejectedUpdate discarded.

6Apply

Apply updates. Write approved changes to the portfolio and refresh the reports.

7Track

Ongoing monitoring and status updates.

02Why AESOP already fits

Reuses existing infrastructure

Same skeleton, same gates

The governance pipeline is the same LangGraph state machine skeleton as the build pipeline. HITL gates reuse the existing interrupt() pattern. Research reuses the Deep Research runner. Document generation reuses the generated_documents table and Supabase Storage.

Net-new work is templates + transcripts

Three renderers and one ingestion path

Three new output renderers (one-pager, pro forma PDF, PPTX deck), one Excel service, and one transcript ingestion + extraction pipeline — all following existing patterns. The transcript pipeline uses the same LLM extraction pattern as the monitor / trace analyzer, with a manual upload fallback so it works without Graph API approval. No new data model, no new pipeline engine, no new approval system.

03Process mapping

Desired process step	AESOP today	Gap
Structured intake form	Discovery interview, `POST /api/projects`, SMB wizard, executive intake	Exists
Validate + capture missing info	Discovery interview gap analysis, triage verdict	Exists
Route to human approver	Working Agreement (draft → sent → signed), HITL checkpoints	Exists
Project research	Deep Research runner, Exa, Serper, SEC Edgar, Crunchbase	Exists
One-page project summary	Opportunity brief compilation, working agreement rendering	New template
Pro forma	Not yet — same Jinja2 render pattern as working agreement	New template
PowerPoint governance deck	Not yet — PDF generation exists via weasyprint	New service
Route for governance review	Build approval gates (approve/reject with comments)	New gate variant
Update portfolio system of record	DISCO portfolio layer: Kanban, scoring, strategic prioritization	Exists
Refresh Excel / PowerPoint reporting	Not yet — CSV export only	New service
Meeting transcript monitoring	Not yet — no transcript integration	New service

04Implementation plan

Phase 1 · 5–7 daysCore pipeline

●Add the governance archetype + DB migration

●Governance intake question bank

●Research node + prompt template

●Material generation node (one-pager + pro forma)

●Wire nodes into the LangGraph graph

●Governance review HITL gate

Phase 2 · 5–7 daysOutputs + transcripts + frontend

●PPTX generation service (python-pptx)

●Excel portfolio report (openpyxl)

●Transcript ingestion (Graph API + manual fallback)

●Transcript extraction LLM pipeline

●Meeting update review HITL gate

●Governance project detail page

●Governance review + meeting update interfaces

●Portfolio governance dashboard view

Humans approve all material decisions. The agent assists, drafts, summarizes, and recommends — but every approval gate, governance decision, financial assumption, budget change, timeline change, scope change, and transcript-extracted status update routes through a human reviewer before anything is formally changed in the portfolio. Same HITL pattern AESOP already enforces in the build pipeline.