I recently interviewed executives from a Palo Alto (US) company called Maana for our sister publication Digital Energy Journal (the full article is here).
Maana doesn’t present itself as ‘low code’ but is certainly a useful platform tool which could be used to make ‘Software for Experts’.
The main focus is helping companies to search and analyse their data – and the target market is large companies which probably have hundreds of databases and no idea how they all connect.
Perhaps a simple way to explain Maana is that it automatically helps to work out the structure of whatever data set it can see, enabling it to come up with conclusions that would be hard or impossible to do manually.
Maana’s search engine crawls, mines, analyses, classifies, clusters, connects and correlates the data, using statistics and machine learning.
Maana can mine datasets with varying degrees of structure or lack of it. For example with structured data, Maana can work out which columns have correlations with which columns, and what might be telling the user something useful.
One of the co-founders is Donald Thompson, who was previously at Microsoft, where he founded Bing’s Knowledge and Reasoning Team (project Satori), and co-founded Microsoft’s project Arena.
Maana could be used as a basis for specialist tools for expert users – combining its search capability with a little custom programming to build a very useful tool.
I talked to Maana about some applications in the oil and gas industry.
It could be used to help avoid oil and gas drilling problems. Most drilling problems come down to physics and rock properties, such as drilling bits getting stuck. And different parts of the world have the same physics and similar rock, yet they don’t share information very much, Mr Dalgliesh says.
The software could help someone answer questions like ‘find me data for when our company had a similar problem to this, drilling in a similar geology to this, and what the company did about the problem.’
If you were getting ready to plan a well, you could use the system to find data about other similar wells, and how the drilling went, and how they produced.
You might notice (for example) that most of the permit requests for this sort of well had to be submitted several times. You can use this information to revise your expected timescales or make sure you get your permit right the first time.
You could work out which decisions made by the drilling department had the biggest impact on production a decade later, which completion techniques yield the fewest production failures, which rigs are the most efficient, which suppliers are the best, where is the best place to do well workovers, and how efficiencies can be found.
The software can be used for predictive analytics – for example one power turbine can generate 15 tb of sensor data a year, which can be used to do predictive analytics.
It could be used in maintenance. If a field technician needs to repair a certain item, Maana can list some of the problems which all those items have had before, and which parts were needed to fix them, so the technician can be make sure she has those parts in her bag.
Adrilling engineer could use the tool to write a ‘classifier’ for kick detection – a tool which would scan various real time drilling data and spot patterns indicating that a kick was happening or about to happen.
Afterwards you could analyse all the kicks which a company had during the year, what was happening before they occurred.
You could look factors such as whether a certain superintendent was on duty when many of them happened.
As another example, you might need to make a difficult decision about whether a certain well is safe for running (vertically suspended) wireline tools, because you are not sure of the gradient of the well (or if the data you have about the gradient is correct). A tool could be written to assess all the available data and make a best estimate.
You can link together different searches in sequence, for example one step to determine if a certain image is a photograph or a drawing, if it’s a photograph have a follow-up step to see if it is a face, if it is a face have a follow-up step to try to recognise the face.