Using BDA - Bedrock Data Automation as an OCR alternative for Pattern 2 and Pattern 3 #42
Description
BDA provides majority of Textract output like, OCR, layout, forms, tables, etc. and even more, like figures, diagrams, captions etc.
It can even provide the description of figures and digrams using Generative AI.
It costs 0.01 cent/page which is much cheaper than Textract if you use all the textract features like layout, forms, tables.
Describe the solution you'd like
Extend Pattern 2 and 3 with BDA standard output.
I see two main approaches since BDA is async only.
a) use sync approach and wait in the lambda until BDA is done -> quick replacement only, and waste of lambda time and costs
b) async usage of BDA by creating simplified version of pattern 1, that gets triggered as prestep after the queue processing. Lets call that step BDA-preprocess-async step. This BDA-preprocess-async step can than be put in front of pattern 2 and 3.
- Results of BDA output will be notified through event bridge, which we redirect into a SQS.
- Benefit is that its async and no payment of lambda is needed when nothing is happening.
- scalles better as previous approach.
Ideally we convert the BDA results into a flat form like a csv which can be also ingested into knowledge bases and utilized for other approaches like few-shot prompting etc.
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context or screenshots about the feature request here.