Skip to content
Business successionPrototype, public demo

LegacyBridge Help the business depend less on its owner.

Independent project built outside my employment.

legacybridge-theta.vercel.app
The LegacyBridge home page. The headline reads, you can retire, your legacy doesn’t have to. Beside it, an exact quote from the fictional owner’s interview, the access codes for their roof are only in my phone, and the dependency LegacyBridge recorded from it.

The problem

In many small businesses, key knowledge lives only with the owner: the supplier contact, the access code, the reason one customer gets special terms. When the owner steps back, that knowledge leaves too.

What I built

An interview tool that asks the owner adaptive questions, records each dependency in the owner’s exact words, and turns confirmed findings into handover tasks for a successor.

What works today

  • Owner interview by voice or typing
  • Each finding keeps an exact quote, checked word for word against what the owner said
  • The owner confirms findings and approves tasks before a successor sees anything
  • Unresolved items carry into the next session

The evidence

  • Public demo, no account needed
  • 181 automated tests passing

The current limits

  • Prototype. The demo business is fictional
  • The Alexa+ voice experience is simulated in the browser. It is not an Alexa integration
  • Decision support only, not legal, tax, valuation, or lending advice
  • Source code is private

How it works

  • Next.js on Vercel with Supabase Postgres. Claude runs only on the server.
  • Every owner message is saved before any AI call, and a database trigger blocks later edits.
  • Claude extracts owner dependencies from the conversation. Each one must carry an exact quote from the owner’s own messages.
  • Prepared demo replies replay analysis recorded from a live session and are labelled as recorded. Answers a visitor types are analyzed live, within usage caps.

Controls

  • Quotes are checked character for character against stored owner messages, in code and again by a database trigger. A finding without a matching quote is discarded.
  • Everything the AI produces starts as unverified. Only the owner can confirm it.
  • A handover task reaches the successor only after the owner confirms the finding and approves the task. The successor never sees the conversation.
  • Every request checks the caller’s role on the server. Row-level security is a second layer.

Hard calls

No quote, no finding

A model summarizing an owner is easy to write and hard to trust. Requiring the exact sentence behind every finding makes each one checkable by the person who said it.

The owner decides what is shared

Handover is personal. Nothing reaches a successor until the owner has confirmed it and approved the task, and the conversation itself is never shared.

Planned, not built

  • A successor interview that shows where the owner’s and successor’s understanding differ
  • Matching owners with operators who share their values
  • Help for owners without heirs to describe what the business should stand for

My contribution

  • Designed and built the product: interview flow, extraction prompts, quote validation, the approval workflow, and the database rules behind it.
  • Wrote the fictional business used in the public demo.
Built withNext.jsTypeScriptSupabaseClaudeVitestVercel