Skip to main content
AI Automation & Agents· Web Application

Guided memorial-writing assistant

A guided writing product that helps families create a personal obituary: a gentle interview, then drafting through a self-correcting LangGraph workflow that scores and revises its own work.

  • Sector Funeral and memorial services
  • Client US memorial-writing product
Draft quality score on the evaluation set
4.8/5
Every draft is evaluated, and weak ones are regenerated
Scored
An interview draws out memories in the family's own words
Guided
Migrated to o3 and re-tuned when a model was retired
Maintained
The starting point

The challenge

Families write obituaries at the hardest possible moment. They need something personal and polished, in their own voice, without an editor, and generic AI text feels cold and wrong.

The solution

What we built

  • Designed a conversational interview that gathers memories, relationships and details step by step, at the family's pace.

  • Built a multi-step LangGraph workflow that drafts, critiques and revises the obituary while preserving the family's own words.

  • Added a self-correcting loop: every draft is scored, and weak ones are regenerated before the family sees them.

  • When OpenAI retired o1-preview, migrated the production graph to o3. The first replacement scored below the bar, so a more detailed regeneration graph was built and deployed.

Architecture

How the system works

How it fits together
  1. 01Guided interviewMemories and details
  2. 02DraftLangGraph writing workflow
  3. 03ScoreEvaluated against criteria
  4. 04RegenerateWeak drafts are rewritten
  5. 05Family reviewEdit, approve and share
Engineering

How it was engineered

  • Scored drafts with evaluation workflows after every iteration and in production, using curated and synthetic family inputs.

  • Traced every run in LangSmith, from interview answers to the final draft.

  • Kept scoring in production after model changes, which is how the first o3 graph's lower quality was spotted and replaced.

Scope

What was delivered

01

Guided interview flow

02

LangGraph drafting and revision workflow

03

Evaluation and scoring workflows

04

Model migration to o3

05

LangSmith tracing

Technology

Stack

LangChainLangGraphLangSmithOpenAI o3FastAPIVercel
Client feedback

In the client’s words

“Always do the right thing, that's why I trust you.”
Client leadUS memorial-writing product · via Slack

Need something similar?

We can scope and build a system with comparable capabilities, shaped around your own workflows, data and integrations.