1. Business Goals
   1. Advise users on interior design choices
   2. Decode the essence of style.
   3. user experience testing
2. Data Engineering Goals
   1. Build initial data pipeline
   2. Figure out how to integrate model with application.
   3. 1. Launch product with heuristics ASAP.
   4. Incorporate ML into the product
      1. monitor performance as a function of "freshness"
      2. monitor feature coverage, watch for silent failures
   5. Test infrastructure independently of ML
      1. Data Ingestion
      2. Exporting trained models
      3. unit testing
      4. system testing
   6. Develop visualzation/debugging tools
   7. Configure automated logging for online processes
3. Data Science Goals
   1. 1. Design and implement heuristics and metrics.
   2. Define what makes a system "good" and "bad"
   3. Consolidate high quality corpuses for ML
   4. 2. Shift from heuristics to simple machine learning models.
      1. Define Objective
         1. don't overthink it
         2. make ML objective simple. easy to measure, and a proxy for the "true" metric
            1. To avoid:
            2. Is the user happy using the product?
            3. Is the user satisfied with the experience?
            4. Is the product improving the user’s overall wellbeing?
      2. Choose simple features
      3. Turn Heuristics into features.
         1. preprocess with the heuristic
         2. directly create a feature
         3. mine raw inputs of heuristic
         4. modify the label
      4. start with an interpretable model
         1. debugging is easier
         2. serves as a baseline
         3. manage feedback loops
         4. decision matters more than the likelihood of the model
      5. Postpone spam considerations, focus on data posted in good-faith.
      6. ensure that model learns reasonable weights
      7. ensure features reach the model correctly
      8. automate "sanity-checks" before deploying models
   5. 3. Iterative improvement of simple ML
      1. Measure first, optimize second
      2. Featurize model errors
   6. Move to more complex models that have been pioneered in the open source community
      1. Transfer Learning
      2. Finetuning
      3. end-to-end training
      4. ensembles