We live on the edge.
Here's how.

Union doesn't guess. It computes. Every introduction is the output of a graph-theoretic system that maps 20 million relationships, scores trust, optimizes paths, and delivers the shortest warm route between you and the person you need.

20M
Relationship nodes indexed
847K
Edges scored per query
3.2
Average hops to target
94%
Warm intro success rate
01 — THE GRAPH

People are nodes.
Relationships are edges.

Every person in Union's system exists as a node in a directed, weighted graph. Every relationship — professional, social, transactional — is an edge with properties: strength, recency, context, and trust score.

This isn't a contact list. It's a living topology of human connection. Edges decay over time. New signals refresh them. The graph breathes.

DIRECTED GRAPH WEIGHTED EDGES TEMPORAL DECAY SIGNAL REFRESH TRUST PROPAGATION
02 — INGESTION

We ingest everything.
We store only signals.

Union continuously scans public professional networks, company registrations, board appointments, funding rounds, speaking engagements, published works, and social signals.

Raw data is ephemeral. What persists is the computed relationship signal — a normalized score that answers: how connected are these two people, and through what context?

INGEST LINKEDIN PUBLIC GRAPH — 2.4M NEW EDGES
INGEST CRUNCHBASE FUNDING Q4 — 847 ROUNDS
COMPUTE EDGE WEIGHTS — BATCH 7B...
EDGE SCORE UPDATED: 14,291 RELATIONSHIPS
INGEST BOARD APPOINTMENTS — 312 NEW NODES
TEMPORAL DECAY PASS — 0.997 FACTOR
SIGNAL REFRESH: 847K EDGES ACTIVE
INGEST CONFERENCE SPEAKERS — SAASTR 2024
DEDUPE PASS — 4,291 MERGED NODES
CLUSTER REBALANCE: 89 COMMUNITIES
INGEST PATENT CO-AUTHORSHIPS — 1.2K EDGES
TRUST PROPAGATION — 3 ITERATIONS
GRAPH CHECKPOINT: v4,847
03 — THE QUERY

You describe a person.
We describe a subgraph.

When you tell Union who you're looking for, we translate your natural language into a set of graph constraints. "A CTO who's scaled a B2B SaaS from 1M to 10M ARR" becomes a subgraph filter: role=CTO, company_stage=growth, revenue_range=[1M,10M], sector=B2B_SAAS.

The candidate set emerges not from keyword matching, but from structural position in the graph. People who occupy similar topological positions to successful matches are weighted higher.

// UNION QUERY PIPELINE INPUT "CTO who scaled B2B SaaS 1M→10M ARR" // STEP 1: SEMANTIC DECOMPOSITION PARSE ROLE="CTO" STAGE="GROWTH" SECTOR="B2B_SAAS" FILTER REVENUE_RANGE=[1000000, 10000000] // STEP 2: TOPOLOGICAL CANDIDATE SET QUERY SUBGRAPH WHERE CONSTRAINTS MATCH EXPAND STRUCTURAL_SIMILARITY RADIUS=2 RESULT CANDIDATES=847 // STEP 3: RELEVANCE SCORING SCORE EACH CANDIDATE × TRUST × RECENCY × CONTEXT RANK TOP 15 BY COMPOSITE SCORE OUTPUT RANKED_CANDIDATES WITH PATH_DATA
04 — PATH OPTIMIZATION

The shortest warm path.
Not the shortest path.

Dijkstra finds the shortest path. Union finds the warmest. Our modified path algorithm weights edges not just by distance, but by trust score, relationship recency, and introduction willingness.

A 4-hop path through trusted colleagues outperforms a 2-hop path through weak acquaintances. Every time.

MODIFIED DIJKSTRA TRUST-WEIGHTED RECENCY BIAS WILLINGNESS SCORE
05 — THE PIPELINE

Five stages. Zero cold outreach.

Every introduction Union makes passes through a five-stage pipeline. Each stage is a gate. If any stage fails the trust threshold, the introduction is rerouted or held.

STAGE 01

Scan

Ingest signals. Index new nodes and edges. Refresh decay scores.

STAGE 02

Model

Translate your request into graph constraints. Build candidate subgraph.

STAGE 03

Score

Rank candidates by composite trust, relevance, and path warmth.

STAGE 04

Route

Compute optimal warm path. Verify each intermediary's willingness.

STAGE 05

Connect

Orchestrate the introduction. Monitor follow-through. Score outcome.

06 — TRUST SCORING

Trust isn't binary.
It's a tensor.

Union's trust model computes a multi-dimensional score for every edge. It's not just "do these people know each other." It's: how well, in what context, how recently, and would they vouch for each other?

Trust propagates through the graph but decays with distance. A first-degree trusted connection carries more weight than a third-degree connection through strangers. The decay curve is exponential, not linear.

Every successful introduction feeds back into the trust model. Every ignored introduction decays it. The system learns.

// TRUST TENSOR COMPONENTS STRENGTH 0.01.0 // interaction frequency RECENCY 0.01.0 // time since last signal CONTEXT [WORK, SOCIAL, BOARD, INVEST] WILLINGNESS 0.01.0 // intro acceptance rate RECIPROCITY 0.01.0 // bidirectional strength // COMPOSITE SCORE TRUST = STRENGTH × RECENCY × WILLINGNESS × RECIPROCITY DECAY = e^(-0.003 × DISTANCE_HOPS) FINAL = TRUST × DECAY × CONTEXT_MATCH
07 — THE RESULT

Not introductions.
The right introductions.

3.2x
Higher response rate vs cold outreach
72hrs
Average time to first meeting
89%
Intros that lead to ongoing relationship
2.1
Average degrees of separation

Ready to see your graph?

Union is computing 847K edges per second. Yours are waiting.

Put me to work