Paper Trail

The documents that run my life today and the documents that preserve my family’s past

Published Voice Engineer Tags Paper TrailAIArchitecture

Paper Trail began with a simple rule: make documents useful without separating them from their source.

Every month I had to pay household bills, collect company documents, and prepare another package for the bookkeeper. I wanted that work to require less of my time and as little paper as possible.

At the same time, my family already had a large archive of medical, legal, property, and historical documents. Some of the property and historical documents were more than a century old. I did not want merely to scan them and forget them on another hard drive. I wanted to understand what they said, make them useful, and exchange the historical material with relatives.

The basic plan was simple: solve one real workflow end to end, keep every original document as the source of truth, and let actual use determine what came next.

The problems worth solving

Current paperwork: household bills and company documents

I deliberately started with mundane paperwork. Household bills and company paperwork arrived continuously, each triggering its own small chain of manual work: review it, act on it, file it, and, when necessary, send it to the bookkeeper.

  • Track each household bill, its due date, amount, and payment status, with proof of payment linked from the relevant bank statement.
  • Organize company invoices and accounting documents.
  • Prepare documents for the bookkeeper with a final human approval step.
  • Reduce paper and repeated manual work without giving the system authority over payments or external communication.

A navigable family medical history

I already had my complete medical documentation, as well as my children’s. The challenge was not collecting the records, but turning years of reports into a private, navigable history that could help my family and our doctors without separating any conclusion from its source.

  • Keep reports, laboratory results, diagnoses, prescriptions and discharge letters in one private, navigable collection.
  • Build a medical timeline for each family member.
  • Ask questions across years of records and receive answers that open the exact original document, page and relevant section.
  • Prepare a structured, source-linked summary before an appointment so a doctor can review years of records without reading every document from the beginning.
  • Surface trends, conflicting information and gaps for the doctor’s attention without presenting generated conclusions as established medical facts.

The legal and property archive spans multiple cases and generations. Some property documents are more than a century old, while the legal records include cases reaching back to the period after the Second World War. No single document contains the whole story; understanding it requires reconstructing timelines across many of them.

  • Extract people, dates, properties, decisions, and references to earlier documents.
  • Reconstruct the timeline of each case or property.
  • Keep every event linked to the source document and page that supports it.

Reconstructing family history

Many of the oldest family documents had survived longer than the context needed to understand them. A name, place, or date that meant little to me might be the missing piece for a relative, and they might hold the missing piece for mine. The goal was not only to digitize the papers, but to rebuild their context without confusing later family interpretations with the original record.

  • Extract people, places, dates, and relationships from century-old documents, with human verification.
  • Connect documents that describe different parts of the same family story.
  • Share scans with relatives and collect names, corrections, and missing context.
  • Keep every extracted fact and later family interpretation linked to its source, while preserving the source document unchanged.

Development plan

Make something usable as soon as possible.

Not a complete archive, not a platform, and not an AI demo: one end-to-end workflow that solved a real problem, followed by the next one.

What was certain

Phase 1: Solve the current paperwork

  • The first end-to-end version handled household bills, company documents, and the bookkeeper workflow.
  • It was already in regular use.
  • It reduced paper and repeated manual work.

Phase 2: Expand collections, one at a time

  • The next step was to choose one existing collection: medical, legal and property, or historical family documents.
  • The plan was to digitize enough of it to create a reliable, usable collection.
  • Every original document would remain the source of truth.
  • The chosen collection had to become searchable and navigable before more documents were added.

Phase 3: Build one real use case at a time

The planned use cases were:

  • Medical history and doctor-assistance workflows.
  • Legal and property timelines.
  • Historical family documents and exchange with relatives.
  • Each use case had to be useful on its own.
  • Real use would determine whether to continue, change direction, or choose the next collection.

What remained open

  • Which collection would come next.
  • Whether I would use one application or several specialized systems.
  • Which technologies, models, and storage would survive the experiments.
  • What the next step would be after each working version entered real use.

Nice to have

  • One shared interface across the different archives.
  • Questions that cross multiple otherwise separate collections.
  • Fully automatic links between related people, events and documents.
  • A portal where relatives can contribute material and context themselves.
  • Advanced timeline, relationship and family-history visualizations.

Conclusion

The first working version covered only a fraction of these possibilities, but that was no reason to design all of them in advance. It already handled real paperwork and was in regular use.

The bills could not wait. The family archive had already waited for more than a century. I started with the immediate problem because it gave me something concrete to solve, while the historical archive deserved a more careful approach.

This post is the map. Each of these problems gets its own entry in the series, one working step at a time, and the links will collect here as they are published.

Paper Trail began with what already worked, not with what a roadmap promised.