[Epic] vNext AI Integration Rollout for BeFam¶
Last reviewed: April 12, 2026
Short conclusion¶
AI can be a standout BeFam feature only when it is embedded into real clan workflows, not shipped as a generic chat tab.
The highest-value AI jobs for BeFam are:
- improving profile quality and recognizability
- drafting event and memorial copy faster
- explaining duplicate-genealogy risk before admins override creation
Firebase AI service assessment¶
Firebase AI Logic¶
Best for:
- low-risk client-side drafting
- lightweight in-app AI moments
- app-level AI monitoring
Not the primary rollout choice for BeFam:
- genealogy data is sensitive
- governance flows need server-side role checks and auditability
Genkit on Firebase Functions¶
Best fit for phase 1:
- server-side control
- structured outputs
- audit logging
- safe fallback behavior
- minimal architecture disruption for the current Flutter + Firebase stack
Firebase Machine Learning¶
Useful later, especially for:
- OCR of scanned genealogy pages
- document digitization
- image preprocessing before data entry
This is more of a phase-2 data-ingestion investment than a phase-1 retention feature.
Sprint rollout¶
Sprint A1 - AI foundation on Firebase Functions¶
Implemented:
- backend AI callable layer in
firebase/functions/src/ai/callables.ts - runtime flags for model selection and safe enablement
- phase-1 default model set to
gemini-2.5-flash-litefor cost control - heuristic fallback when AI is unavailable or not configured
Sprint A2 - AI profile quality reviewer¶
Implemented:
reviewProfileDraftAi- profile editor integration
- structured summary, strengths, missing items, risks, and next steps
Sprint A3 - AI event copy drafting¶
Implemented:
draftEventCopyAi- event editor integration
- suggested title, description, and reminder timing
Sprint A4 - AI duplicate genealogy explanation¶
Implemented:
explainDuplicateGenealogyAi- duplicate-risk explanation before override
- review checklist for admins
Why this can become a standout feature¶
Yes, if BeFam keeps AI:
- task-specific
- short and actionable
- advisory only
- inside existing workflows
No, if AI becomes:
- a generic chat surface
- auto-committing data
- detached from the user’s current task
Retention impact¶
The strongest retention upside is likely for:
- profile completion
- memorial/event operations
- admin genealogy setup flows
These are higher-intent moments than a standalone assistant surface.
Required guardrails¶
- AI suggestions remain advisory
- sensitive writes still require explicit confirmation
- server-side role checks remain authoritative
- every AI flow keeps a safe fallback path
Recommended next phase¶
- Vietnamese natural-language family search
- OCR-assisted genealogy record digitization
- relationship suggestion assistant for admin review