CRM
Entity resolution
One person appears on seven platforms with different identifiers. A 4-phase probabilistic pipeline unifies them into a single entity.
The problem
Same person, seven platforms
Identifiers
Gmailjohn@company.com
Slack@john.smith
WhatsApp+1-555-0123
iMessage+1-555-0123
MeetingJohn Smith (attendee)
Telegram@john_s
Outlookj.smith@company.com
Resolved entity
John Smith
7 aliases across 4 active channels
156 interactions tracked
Pipeline
4-phase resolution waterfall
Each phase is progressively more expensive. Fast exact matches resolve first. AI-assisted fuzzy matching handles edge cases.
1
Alias lookup
Exact match on known aliases. Email, phone, username.
O(1) hash lookup, ~0.5ms
2
Semantic match
384-dimensional embeddings with cosine similarity threshold.
Vector search, ~5ms
3
Fuzzy + AI scoring
Phonetic matching combined with Fellegi-Sunter probabilistic scoring. AI confirmation for uncertain cases.
~50-500ms
4
Entity creation
New entity created with initial alias registration when no match found.
DB write, ~10ms
Resolution flow
Incoming Identifier
(email, phone, name, username)
|
Phase 1: Alias Lookup
Exact match on known aliases ──── found? ── return entity
|
| not found
|
Phase 2: Semantic Match
384-dim embeddings, cosine sim ── match? ── return entity
|
| no match
|
Phase 3: Fuzzy + AI
Phonetic match + Fellegi-Sunter
> 0.85 ── auto-merge
0.6-0.85 ── AI confirmation
< 0.6 ── separate entities
|
| no match
|
Phase 4: Create
New entity + initial aliasScoring
Fellegi-Sunter probabilistic model
Each identifier field contributes a likelihood ratio. The composite score determines automatic merge, AI review, or separation.
| Field | Weight | P(match) | P(non-match) | Ratio |
|---|---|---|---|---|
| Email (exact) | High | 0.95 | 0.01 | 95.0 |
| Phone (E.164) | High | 0.90 | 0.02 | 45.0 |
| Name + Org | Medium | 0.80 | 0.10 | 8.0 |
| Name only | Low | 0.60 | 0.25 | 2.4 |
| Username | Low | 0.50 | 0.30 | 1.7 |
≥ 0.85
Auto-merge
0.6 \u2013 0.85
AI review
< 0.6
Separate entities
Intelligence
Entity intelligence
Deep context builds
7-phase pipeline producing comprehensive research briefs from all available interaction data.
Entity briefing engine
Personalized briefings per CRM section. Communication style, key topics, relationship history.
Interaction ledger
Cross-channel communication history with frequency tracking and recency scoring.
Relationship tiers
Automated health scoring based on interaction frequency, depth, and recency.
Co-occurrence analysis
Tracks who appears together across meetings, emails, and group conversations.
Outreach suggestions
Warmup recommendations when relationship health decays below threshold.
Architecture
7 focused modules
Module structure
universal-resolver.ts (entry point) alias-lookup.ts match-strategies.ts probabilistic-scorer.ts entity-writer.ts email-utils.ts batch-resolver.ts entity-audit-engine.ts
0Resolution Phases
0Embedding Dims
0Modules
Talk to Sentigen
Have questions about what you're reading? The intelligence behind this platform can walk you through it.