ThinkTopic with Jeff Rose founder and CEO.
ThinkTopic is a tech firm in Boulder, Colorado that was founded a short two years ago combining the latest discoveries of machine learning with expertise and systems engineering. Rather than a typical data science company that might produce some reports or generate some insights, ThinkTopic builds tools and systems that are meant to empower the business owner.
…is a machine learning-focused software consultancy technology company that works in a broad range of different types of problems, some computer vision, data science, signal processing, doing some work in biotech, financial tech, kind of real wide range of areas.
…when you create a machine learning model, in essence what you’re doing is writing a program that’s taking in some data and adjusting its parameters based on what data it’s been exposed to so far, with the goal of sort of modeling that data to kind of understand it. Meaning what are the set of states that data can take and what do those states mean.
…is excelling at, tackling fresh problems that businesses or organizations are facing and they’re maybe not quite sure how to approach it or what types of techniques would be appropriate or how hard of a problem is it even. Is it feasible at all?
…diagnostic development and in using neural networks and machine learning models to analyze data coming from some of their proteomic analysis technology. So in essence we get results from a very sophisticated blood test that shows the set of concentrations of various molecules in the bloodstream. And then we’re working on developing models to help make diagnostic decisions or predictions or sort of score people based on that information.
…so for that project I worked with a GIS consultant and what we did is created a tool where we licensed data from a number of data provider companies, also the Department of Energy, found a bunch of open data, and built a tool where they could, in essence, select Tajikistan and say, “Okay, show us the regions that have at least 10 meters per second of average wind speed at an 80-meter hub height.”
…they could create a constraint like that, and then hit “enter” and it would show the sweet spots on a map. And then they could add an additional constraint and say, “I must be less than this slope and it must be within a certain distance to a road and within a certain distance to a power line and have less than a certain population density,” and so on and so on. And then they could really narrow into the key spots to then actually fly out and investigate.
Cell Phone Network
…we were trying to look at usage data from their users in terms of how many calls were they making, how many are getting dropped, are they calling people in or out of network, in-country or out-of-country, potentially even the web of phone numbers that they’re calling, and try to design some algorithms that could help the company to know who is likely to leave next month.
… it’s a fairly straightforward exercise to build a model that can take in a data set like that, predict who’s likely to leave given historical data, and then you can train another model that actually learns over time which type of offer is this person likely to benefit most from or likely to stick around because they were offered it.
…I think that’s something that particularly my brother is great at celebrating, like when we get new stuff working. And in software we’re building things that kind of have lots of moving parts and it’s exciting when you finally turn on a new system and it does the right thing. We really try to celebrate our creations.
How to reach Jeff-