Thought Leaders in Big Data: Ganes Kesari, Cofounder & Chief Decision Scientist of Gramener (Part 3)

Sramana Mitra: You don’t use off-the-shelf visualization tools like Tableau or FusionCharts? There’s a whole bunch of visualization capabilities available. You don’t use any of that. You do everything from scratch.

Ganes Kesari: That’s right. We don’t use the official product. The approach we follow is more programmatic. Our platform is built on Python and use Javascript. It integrates with D3 charts.

The reason is, we believe there is a lot more flexibility. For someone who understands data and design, you take this approach and it gives you a lot more flexibility to build powerful visualizations which may not be possible while using something like Excel or Tableau.

Building a narrative with a limited set is going to get very difficult. That is why having an open platform is useful.

Sramana Mitra: I’m surprised that nobody has a product that can fill that gap. What you’re identifying is a gap in the market where there should be a product.

Ganes Kesari: Right. D3 is fairly popular. The barrier is, it has a steep learning curve. You need people skilled in Javascript and Python. You’re right. On one side, you have drag and drop but with limited capability. On the other hand, you have programmatic visualizations, which gives you enormous flexibility but calls for certain skills.

Sramana Mitra: Let’s talk about another couple of use cases so that I get a flavor of the scope of your work. You answered one of my interesting questions which are gaps in the market where people should be developing products already.

Ganes Kesari: There are a couple of other use cases. I’ll talk about a completely different area where we’re using the deep learning model. We have a pharma customer where they were facing challenges in the drug discovery process. They were trying to simplify the drug characterization.

In simple terms, you have a microscopic image and you are trying to count the number of cells of different shapes. You need to count and identify the areas of these cells and the characterization of the drug that tells certain details about the drug.

They do this for every drug and every trial. This is a tedious manual process. There are pathologists who take the picture and sit and count. We applied crowd-counting algorithms. We extended that approach and applied it here to count the different shapes of cells with a very good level of accuracy.

That’s stage one where we applied a highly-engineered deep learning algorithm. The second one was bringing it into the workflow. How do we build pollution into a user’s workflow so that adoption happens. Otherwise there are projects that have failure rates of 80%.

Technically, the problem is solved, but it is not put into the business workflow. We built it into the workflow where the pathologist will just upload the image and the algorithm presents the result. They have the ability to go in and inspect. We built our visualization layer in this as well. That’s a classic case of AI and visualization solving a core business problem.

Sramana Mitra: You’re doing this all from scratch? 

Ganes Kesari: Based on the user, we do a white-boarding of what is the right workflow for the user. What representations are relevant? For more technical users, they can have more interactivity. We customize layering and the extent of detail and narrative.

This segment is part 3 in the series : Thought Leaders in Big Data: Ganes Kesari, Cofounder & Chief Decision Scientist of Gramener
1 2 3 4