Network Intelligence System
Unified architecture for Getro's network intelligence: connection paths (spec 007), relationship strength + reachability + key connection (spec 008). Built on shared Sidekiq + Postgres caching infrastructure with Findem enrichment.
TL;DR
Two intelligence layers, one infrastructure. Spec 007 surfaces who can intro you to companies on your lists (direct + work-overlap intro paths). Spec 008 turns email + calendar metadata into Warm/Known/Cold strength tiers, reachability rollups, and key-connection picks. Both share the same per-pair caches, edge tables, and rollup architecture documented in the technical spec. 007 v1 ships independently of email/calendar ingestion; 008 thin V1 ships at ~50% heuristic coverage and grows to ~88% as Findem enrichment lands.
Unified architecture & execution plan
For engineers. The single source of truth for the system — services, data model, caching, integrations, phased plan covering both 007 and 008.
- System architecture + ER diagrams
- Data model: per-pair caches + edges + rollups
- §6 Caching architecture (English + technical)
- Rule engine & scoring service
- Operational envelope (numbers)
- Phased plan covering 007 v1/v2/v3 + 008 slices
- Decision records — DR-01 through ADR-007-B
From data to strength
For PM, design, reviewers. What data we ingest and how it becomes a signal.
- Raw data → signal primitives
- Every heuristic in plain English + code
- Combining clauses into tiers
- Reachability rollup
- Worked scenarios + scale walkthrough
- Privacy guardrails
Google & Microsoft — per provider
For backend engineers. What exists, what extends, what's greenfield, with code skeletons and official doc links.
- Verified current-state audit
- OAuth flow + scope management
- Gmail metadata + Calendar clients
- MS Graph Mail + Calendar (greenfield)
- Azure AD multi-tenant setup
- Shared-mailbox decision matrix
Graph DB vs Postgres for connection paths
For tech leads + reviewers. The long-form analysis behind ADR-007-A — why we chose Postgres tables over Neo4j, Apache AGE, and live JOINs.
- The two query patterns (list view + drill-in)
- Why this is not deeply graph-shaped
- Options matrix & performance comparison
- SQL vs Cypher side-by-side
- Multi-edge-type schema strategy (5 path types)
- Operational cost breakdown + revisit triggers
Findem code findings (firstcut + app-next)
For engineers + reviewers. What Findem already does, with clickable links to source files in firstcut and app-next.
- The overlap kernel (formulas worth copying)
- connection_svc batch path + 15k cap
- LoadConnections runtime merge (backend + app-next consumer)
- sandbox/matches — what it does and doesn't include
- Profile data model (sparse connections, missing score)
- Findem capability gaps F1–F9