This is a serverless photo album web application, that can be searched using natural language through both text and voice. The architecture diagram is as follow:
- B1 is a S3 bucket stores front-end stuff, together with API Gateway, LF2 (Lambda Function 2) forming a website that can handle user request (upload photo, search photo by text, search photo by voice, etc.)
- A new photo is uploaded to another S3 bucket B2, which that trigger LF1 to parse the photo using AWS Rekognition that returns some keywords of the photo. Then LF1 stores the keywords with storage information of the photo in the ElasticSearch for furture indexing.
- When user types in a sentence like "show me my cats", it is handled by LF2, which further utilize Lex to parse the sentence and get the keyword ('cats' in this case), then return the corresponding photos that contains that keyword.
A cloudformation template yaml file is in the root directory, which automatically builds up all the necessary Amazon Web Services, including two pipelines that help automatic code update from github and continuous delivery.
- LFPipeline: Build two lambda functions and the S3 storing photos (there's a trigger so I put them together for convenience). Plus, api gateway is also built here.
- PipelineS3: This is for the front-end code. Note that the front-end code is in the
front-endbranch - Elastic Search
Note that there's no template for Lex so it should be built manually.