Back to projects

Whakapapa

Preserving family stories with AI before they're lost

2025 — Present
Solo Designer & Developer
Live siteSource
Next.js 16SupabaseClaude APITesseract.jsReact FlowdagreFramer Motion

Ownership

Solo Designer & Developer

Team

Solo / small team

Key Result

MIT Open Source

Overview

An AI-powered family knowledge base that turns old documents, photos, and spoken stories into a living, navigable family tree. Named after the Māori word for genealogy — because ancestry isn't just names and dates, it's the web of stories that connect us.

The Challenge

Every family has stories slipping away. Letters in boxes, photos without names on the back, grandparents whose memories are fading. Existing genealogy tools (Ancestry.com, MyHeritage) are expensive, clunky, and treat family history as data entry. The challenge: build something that makes preservation effortless — AI does the heavy lifting, you just feed it documents and tell it stories.

Constraints

  • Trust is fragile with family history data, so AI output must stay reviewable and reversible.
  • Many source artifacts are low quality scans and handwritten notes.
  • The product needed to be viable for a solo builder and affordable for families.

Decision Log

Problem

AI extraction can hallucinate relationships.

Decision

Added a suggestion queue with explicit human approval before writing to the tree.

Tradeoff

Slightly slower ingestion versus fully automatic sync.

Impact

Higher confidence in accepted data and fewer correction loops.

Problem

Genealogy tools often bury stories under forms.

Decision

Made voice capture and story context first-class alongside people and dates.

Tradeoff

More schema complexity and moderation edge cases.

Impact

Richer family context and stronger emotional retention.

Approach

1. AI Extraction Pipeline

The core innovation: scan a document with your phone camera → Tesseract.js OCR extracts text client-side → Claude AI analyzes the text and returns structured genealogical data (people, dates, relationships, places) → suggestions queue for human review → approved data populates the tree. No manual data entry. The AI handles birth certificates, old letters, newspaper clippings, and handwritten notes.

2. Voice-First Story Capture

Not everything is written down. The app includes voice recording with transcription — sit with a grandparent, hit record, and their story is captured in their own voice. Attached to the people it's about. Interview prompts help guide the conversation when you're not sure what to ask.

3. Living Tree, Not Dead Database

The family tree uses dagre for hierarchical layout and React Flow for interactive visualization. Color-coded relationships (parent-child, partners, siblings), zoom, search, keyboard navigation. But the tree is just one view — memories, stories, photos, and documents are all connected to the people they belong to. It's a knowledge base, not a chart.

4. Open & Respectful

Open source (MIT) and free. Self-hostable. Privacy-first with Supabase Row Level Security. The name honors te ao Māori — whakapapa is a concept where ancestry is a living story, not a static record. That philosophy shaped every design decision.

Outcome

A fully functional genealogy platform that replaces manual data entry with AI extraction. Open source with a growing community. Features that commercial tools charge $20-50/month for — available free.

MIT

Open Source

Doc → OCR → Claude → Tree

AI Pipeline

$0 vs $240/yr

Cost vs Ancestry

Learnings

  • AI extraction isn't magic — it needs a human review step. The suggestion queue is the trust mechanism that makes AI-generated data feel safe to accept.
  • Voice recording changes the emotional register of a genealogy app. Reading a transcription is information; hearing grandma's voice is preservation.
  • Naming matters. 'Whakapapa' immediately communicates that this isn't another Silicon Valley genealogy SaaS — it's built with cultural awareness and respect.
  • Supabase + Next.js is an incredibly productive stack for solo builders. RLS policies handle multi-tenancy without a dedicated backend.

Anti-Patterns Avoided

  • ×Auto-trusting AI inserts directly into canonical records.
  • ×Forcing users into rigid, spreadsheet-like genealogy workflows.
  • ×Locking export/import behind paid plans.

Next Iterations

  • Confidence scoring and source-level reliability badges per extracted entity.
  • Guided interview flows optimized for elder story capture sessions.

Need This Type Of Outcome?

I run focused design engineering sprints for product teams.

If your team needs clearer UX, faster implementation, and measurable performance improvements, I can help.

Next project

Liner