Contacts manager: standalone view with dedup/merge strategy #73

New issue

Closed

opened 2026-04-06 10:05:44 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-04-06 10:05:44 -07:00

Owner

Context

Peregrine already captures contact data implicitly (recruiters, hiring managers, interviewers) through applications and calendar events. This issue gives contacts a proper home with dedup from the start rather than retrofitting later.

UI

Standalone Contacts view — first-class nav item. Contacts outlive individual applications (same recruiter, multiple roles; same hiring manager across companies) and need a dedicated home.

Contact card fields:

Name, title, company, email, phone, LinkedIn URL
Source (where captured: application, calendar event, manual)
Relationship history: linked applications, conversations, interviews
Notes

Dedup signal stack (cheapest to most expensive)

1. Exact email match         → definite duplicate — auto-merge
2. Exact LinkedIn URL match  → definite duplicate — auto-merge
3. Same name + same company  → probable — flag for user review
4. Fuzzy name + same domain  → possible — low-confidence flag
5. LLM similarity score      → edge cases only

Merge strategy

Field-level merge with user confirmation on conflicts.

Non-conflicting fields: fill gaps automatically (most recent non-null wins)
Conflicting fields: surface side-by-side for user to pick canonical value
Never auto-resolve conflicts silently — job search context makes misidentification genuinely harmful

Pipeline crystallization hook

Early ambiguous merges go through LLM-assisted review. User approvals crystallize into deterministic dedup rules over time (Circuit-Forge/circuitforge-core#33).

Acceptance criteria

Contacts view in nav
Contact card with full field set
Linked applications + conversation history per contact
Dedup runs on import and on manual add
Merge confirmation UI for conflicting fields
Manual merge trigger (user selects two contacts → merge flow)
Export contacts (vCard / CSV)

## Context Peregrine already captures contact data implicitly (recruiters, hiring managers, interviewers) through applications and calendar events. This issue gives contacts a proper home with dedup from the start rather than retrofitting later. ## UI Standalone **Contacts** view — first-class nav item. Contacts outlive individual applications (same recruiter, multiple roles; same hiring manager across companies) and need a dedicated home. **Contact card fields:** - Name, title, company, email, phone, LinkedIn URL - Source (where captured: application, calendar event, manual) - Relationship history: linked applications, conversations, interviews - Notes ## Dedup signal stack (cheapest to most expensive) ``` 1. Exact email match → definite duplicate — auto-merge 2. Exact LinkedIn URL match → definite duplicate — auto-merge 3. Same name + same company → probable — flag for user review 4. Fuzzy name + same domain → possible — low-confidence flag 5. LLM similarity score → edge cases only ``` ## Merge strategy **Field-level merge with user confirmation on conflicts.** - Non-conflicting fields: fill gaps automatically (most recent non-null wins) - Conflicting fields: surface side-by-side for user to pick canonical value - Never auto-resolve conflicts silently — job search context makes misidentification genuinely harmful ## Pipeline crystallization hook Early ambiguous merges go through LLM-assisted review. User approvals crystallize into deterministic dedup rules over time (Circuit-Forge/circuitforge-core#33). ## Acceptance criteria - [ ] Contacts view in nav - [ ] Contact card with full field set - [ ] Linked applications + conversation history per contact - [ ] Dedup runs on import and on manual add - [ ] Merge confirmation UI for conflicting fields - [ ] Manual merge trigger (user selects two contacts → merge flow) - [ ] Export contacts (vCard / CSV)