Aishwarya Srinivasan

From a hardware perspective I count database to be the biggest drivers. Having a datalake with access approval pipeline for tailored data access is the most important pillar.

This is wirh respect to reporting tools where one would like to access real-time data to visualize trends. Few of the bottlenecks I have seen in dashboards I have used, which has required me to instead query tables using SQL/Python are: - When I need to pull data from different tables and use relevant joins - The dashboards don't reflect the data freshness I need - There are gaps in access of tables in the dashboards whereas I can access the backend data - The dashboard has predefined set of charts which limits me from visualization what I need to - When I need to add filters/operations to the data to look for something particular or tranpose the data and the dashboard doesn't have those functionalities

As I mentioned in the previous section, navigating through these standards and frameworks can be tricky. But thanks to the researchers in the field, there are different metrics and standards available to us that helps us define this. One of the best toolkits I have used is RAI Toolbox developed by MIT as opensource: https://github.com/mit-ll-responsible-ai/responsible-ai-toolbox

It is important that organizations have a defined process of screening process that has a AI Ethics committee that evaluates any product or solution being built that uses AI in any form. This is where we can have standards for each of the Responsible AI pillars. There are challenges to standardize categories like bias, fairness, even explainability, but there research work available from academia, that will help you through it. Google has such a standard process to evaluate any application that uses AI. You can read more about it here:https://fortune.com/2022/01/11/google-cloud-andrew-moore-ai-ethics-lemonade-iced-tea/

Oftentimes, out of the box data profiling methods don't work as the data has context to it, which requires input from subject matter experts to define some of these rules. There rules can be heavily reliant on exploratory data analysis.

One of the ways I keep up to date with RL methods is by following startup and products being developed by companies. These research paper and blogs being published by their teams are really technical and a great resource to upkill on real-world applications of RL. Here is a list of some companies working on RL and imitation learning: https://www.ventureradar.com/keyword/Reinforcement%20learning Most of their websites will have resources. You can shortlist the companies based on the industry and usecase you want to learn about and check out their resources.