Architecture

A research platform built around the investment workflow.

Aperture replaces fragmented Excel and email workflows with a single, governed pipeline — from raw filings to portfolio-level conviction.

Five-layer architecture

Sources
  • Postgres warehouse (read-only)
  • DuckDB analytical engine
  • Bloomberg DLWS
  • Email & document ingest
Pipelines
  • Document parsing & OCR
  • Earnings normalization
  • Entity resolution
  • Sentiment classifier
Intelligence
  • AI note drafting
  • Thesis summarization
  • Quant factor models
  • Anomaly detection
Surfaces
  • Dashboards
  • Portfolios & holdings
  • Earnings calendar
  • Quant screening
Governance
  • RBAC & access reviews
  • Immutable audit log
  • Data lineage
  • SSO + 2FA

Document → Insight flow

Step 1
Capture
Filings, transcripts, emails
Step 2
Normalize
Parse, OCR, entity resolve
Step 3
Enrich
Sentiment, factor scoring
Step 4
Reason
AI-assisted thesis drafts
Step 5
Decide
Portfolio fit & risk
Step 6
Audit
Immutable trail
Rules engine

Declarative YAML rules evaluated on every ingest event. Each rule can emit notifications, label documents, or trigger compliance workflows.

rule: earnings_miss
when:
  event: earnings.reported
  expr: eps_actual < eps_estimate * 0.95
then:
  - notify: pm.coverage(ticker)
  - label: doc, "Negative surprise"
  - audit: severity=warn
RBAC matrix
CapabilityAnalystPMComplianceAdmin
DocumentsRWRRR
NotesRWRRR
PortfoliosRWRR
Audit logRWR
Users & rolesRW