Internal/admin tools; in-house ground-truth training client.
All critical functions, cron, data-accept -> dynamodb. Light routing.
Sparingly. Gateway to non-scaled components.
cache, ETL activities - roll-up. Anything speed-worthy.
Too much to list. Long-term, short-term, etc. Some event firing.
Public-facing image store.
heavy-lifting CV. Re-written C++ libraries for lambda - MSER, text transform, etc.
basic image processing, transform wrapper.
Here & there. All on lambda. Anything guaranteed cold spin-up; otherwise bit slow (compared to node/jvm) when warm.
Low stress ETLs, light master data.
main data collector -> feeds to ETLs, fires select events.
Cloudwatch monitoring & events for chron-> lambda.
CV glue. Modified libraries for pattern-detection. Some pattern training tasks. HoG matching. Transform
Pilot integration for retraining.
Mild re-training data usage.
Modified tesseract libraries for transform & OCR
Cron data-roll-up to ETL