Create process / infra diagram for scanning

[grossir]

Once raw scans are uploaded to S3 (either phone scanning or Hein data) we intend to perform the following steps

Blackletter service section:

• get raw reporters from S3
• separate opinions out of the reporter file
• censor copyrighted material using YOLO
• save ready to extract opinions to S3

Courtlistener section

• call LLM services to extract XML
• save extracted content and metadata into the DB

So, we have some open questions which fall on Chaco's expertise

• How will black letter be deployed?
• How will it run? How will it know what to process next?
• What hardware does it need to run?
  • is YOLO usage heavy on processor / RAM?

Am I missing anything?

[grossir]

Some updates:

• there will be daemon for blackletter processing
• there will probably be another one (maybe in the same deployment) for blackletter validation, which provides feedback to uploaded PDFs, looking for blurry, duplicated pages or other errors