Cross-Platform Entity Resolution: Deduplicating SaaS Identities

In enterprise data engineering, a major challenge when constructing a corporate knowledge graph is identity fragmentation.

A single senior engineer might exist across SaaS applications as:

• @alex_sf in Slack
• alex-chen-code on GitHub
• alex.chen@company.com in Jira
• Alex Chen in Zoom meeting transcripts

If an enterprise AI knowledge graph treats these representations as four separate people, search queries cannot accurately aggregate contributor context or trace decisions across systems.

In this systems engineering breakdown, we explore common techniques for Cross-Platform Entity Resolution: merging fragmented user profiles into unified contributor nodes. It is the kind of problem any organizational memory platform, including Memora, has to handle to link work across tools.

3-Layer Identity Resolution Pipeline

┌─────────────────────────────────────────────────────────────────────────┐
│                 3-LAYER IDENTITY RESOLUTION PIPELINE                    │
├─────────────────────────────────────────────────────────────────────────┤
│ Layer 1: Deterministic Email & Account ID Match                        │
├─────────────────────────────────────────────────────────────────────────┤
│ Layer 2: Probabilistic String Similarity (Jaro-Winkler & Levenshtein)   │
├─────────────────────────────────────────────────────────────────────────┤
│ Layer 3: Co-Occurrence Graph Topology Clustering                       │
└─────────────────────────────────────────────────────────────────────────┘

Layer 1: Deterministic Email & SSO Matching

The simplest signal is a shared identifier. If a user's GitHub account and Jira account share the same corporate email (alex.chen@company.com), the nodes can be merged with high confidence.

Layer 2: Probabilistic Name Similarity

When email handles are not available (e.g., matching a Slack username to a Git commit author name), a resolver can fall back to string similarity measures such as Jaro-Winkler:

def jaro_winkler_similarity(s1: str, s2: str) -> float:
    # Measures string edit distance weighted by matching prefix
    pass

Layer 3: Co-Occurrence Graph Topology Clustering

If two user nodes frequently co-occur in the same pull requests and Slack threads, graph clustering algorithms increase their identity merge probability score.

Related Technical Resources

• How AI Builds Organizational Memory
• Entity Extraction Schemas for Technical Codebases
• Graph RAG vs Vector RAG Comparison