Conveyor

Files (bucket)

The problem

A home equity operation goes through many hands (partner, analysis, legal, payment) and every stage requires documents and checks. Without a single place, operations stalled and nobody knew which stage they were in or what was missing to move on.

bucket keeps the metadata of every file and never touches the bytes: the browser uploads and downloads straight from Cloudflare R2 through presigned URLs. Public pictures and private documents are separate paths with their own rules.

Decisions and why

Bytes never pass through the API
A 20 MB upload takes no memory or thread from the service.
A document is private for what it is, not its type
A photo of an ID card is a document. The purpose picks the bucket.
Confirm the upload
After the upload the service checks on R2 that the size matches before marking it READY; the checksum is already signed in the URL, so R2 refuses different bytes.
R2 instead of S3
No egress fees: downloading a document many times does not become a bill.
requestresponse
Pick a scenario above to watch the message travel through the architecture.

What it looks like

Replica of the pipeline board with fictional data (internal system)
esteira.evilrabbit.com.br/lines/triage
EsteiraPTGM
Triagem
Conferência de dados2
Marina SouzaGMR$ 480 mil 1 d
Paulo HenriqueR$ 1,2 mi hoje
Combate à fraude1
Ricardo AlvesAPR$ 750 mil 4 d
Validação dos documentos2
Carla PradoGMR$ 610 mil 2 6 d
João ReisR$ 295 mil 2 d
Aguardando aprovação0
Nada aqui