Our software requires manual transcription of images. We are looking to leverage Amazon MTurk for the same. Here's the flow,
1) When an image needs to be transcribed, its stored in amazon S3 bucket. A message is posted to the Amazon SQS with the key of the image.
2) A listener listens for messages in the SQS Queue.
3) When a message is available,
a. The image is retrieved from S3.
b. Using MTurk API, a HIT is created for that image.
c. Using MTurk API, a batch is posted for that HIT requiring two workers to do the task.
d. A notification is also set in MTurk to watch for HIT completion. (using Amazon SQS).
e. Using MTurk API, once HIT is complete, the results from the batch are checked.
f. If identical, the HIT is approved, the workers are paid and the results are written as JSON to Amazon S3.
g. If not identical, the HIT is approved, the workers are paid and an email is sent to a support email address.
None of the above steps are manual....