How cannabis and tomatoes are grown using machine learning



This week, Valeria Kogan, a graduate of Physics and Technology , co-founder of startups Fermata and Smartomica, spoke on our youtube channel .



Lera got the idea to control plants in greenhouses using machine learning when her friends told her about their problems with the mass cultivation of cucumbers and tomatoes. Then she and her friends founded Fermata and started developing a platform for real-time plant monitoring.



In 2019, the company attracted $ 1.1 million in investments from a private investor, and already in March 2020, during round A, it received another $ 3.7 million in investments from the British fund Massa Innovations and several private investors.



In addition to agrotech, Lera is engaged in the development of new methods for diagnosing cancer and is a visiting scientist at Roswell Park Cancer Institute. At Smartomica, they develop technologies for the analysis of medical and scientific data for the diagnosis and treatment of cancer patients. We



share with you the transcript and recording of the broadcast.






My name is Valeria Kogan, I am a co-founder of startups Fermata and Smartnomica. And in connection with the name of this stream, I must make a statement: Fermata, its employees and shareholders do not grow, do not use or recommend prohibited substances. We are engaged only in the analysis of the state of plants, incl. medical cannabis. We analyze and control it on the territory of Israel, where it is legal, and we will never do it where it is illegal.



I will tell you a little about myself, about the projects, where they came from, and I will be happy to answer questions.



I graduated from Physics and Technology, Faculty of Biological and Medical Physics. All the time I was studying at the university, I was engaged in scientific research and really wanted to develop in this direction.



After university, I entered the graduate school of Ariel University. During my studies at the university, and later, I was all the time engaged in data analysis in application to biological and medical problems. At university, I was engaged in bioinformatics, various other tasks related to AI, worked in several startups. But this has always been a purely scientific activity, not related to the practical, rapid application of AI and data analysis to real problems. At the university, when I was already in graduate school, I began to engage with my supervisor in such things that were already a practical application of AI - this later grew into the Smartnomica project. We began to try to apply data analysis, machine learning in order to diagnose cancer patients, choose the right treatment for them, see howwhat I can do can be useful in this area of ​​knowledge.



And then, by chance, an idea arose: although I am a person absolutely far from agriculture, who has never seen vegetables grow - in the course of communication with mutual acquaintances, that what we do for cancer diagnostics can be used for plant diagnostics. And from this thought, from communication with the producers of tomatoes and cucumbers, the idea of ​​Fermata arose. I returned to Moscow, met with people who said that it would be cool to apply AI to agriculture. Then it seemed that not many people do it. They talked about how AI can be useful, what problems they have. And in the course of conversations with them, it became clear that the task of monitoring the state of plants is very urgent. It is solved by people - not very effectively, because the greenhouses are huge, there are a lot of plants and sometimes every leaf needs to be monitored. AND,if a disease is noticed late, then the entire greenhouse may suffer, and then the farmer will suffer losses.



We thought it was a very cool problem - together with my friends, whom I invited to solve it. And we thought that if we now solve the problem of crop loss due to infections with the help of AI, then we will be able to scale this solution to the entire agricultural market. We began to think about how to implement this business project; first - with those people with whom the idea was initially discussed. We did not agree with them, and decided that we would start acting independently and return to them with a finished product. This is how our great adventure story began. We began to form a project and try to get support, funding, understanding that someone needs it. This was not a success story for a long time; we faced many difficult tasks. Firstly, there was no one on our team who understood anything in agriculture:there were only my friends from physics and technology, specialists in AI.



Initially, we wanted to create an agricultural robot - it seemed to us that it was a smart idea, very technological and cool, and that we would be paid a lot of money for a robot. It quickly turned out that this was not the case, but initially we proposed exactly the idea of ​​a robot and wanted to get at least some confirmation that this was a useful and necessary project. We applied to the Bortnik Foundation to get the first funding for the project; I already had a successful student experience with the foundation, and I thought that everything would work out - they would tell us that everything was great, we would get money, and everything would work for us. But instead we did not receive any feedback, they did not give us money, and no one answered the calls. We were very tense and did not understand what this all meant - we had a bad project, or were they simply misunderstood?



Then we tried to contact Skolkovo, contacted the manager in charge of Skolkovo-Agro. We were told that the project is strange, and they do not like it, and that some IT specialists constantly come to them, who think that they understand something in agriculture, and who offer a "spherical horse in a vacuum." As we now understand, this was the correct point of view. When we said that we did not have an agronomist in our team, we were told that further conversation did not make sense. That is, we did not receive support in Skolkovo either.



What to do next, while staying in Russia, we did not understand, and were at a loss. There weren't many incubators then, there was no understanding of where to go. Part of the team had already been assembled, some people worked for free, together with another co-founder, Anton, we worked in our free time and created this product. We already had an AI prototype of a plant disease classifier based on images that we found in the public domain. But there was no financial and moral support. We decided it was a bad symptom; the various people we spoke to in the agricultural industry reacted negatively to the idea of ​​AI.



We decided to make the last attempt to get feedback on the project and submitted an application for the competition of agricultural startups, which was held in Belgorod - "Startup: land". We almost missed the train, but in the end we got on it - and won, to our surprise. There were many interesting projects from people experienced in agriculture, and there was a wonderful jury with experts who gave great feedback to our work. As a result, it was there that people very much supported us - even the Skolkovo representatives who were there. We won the tender, and formally this was supposed to ensure our implementation in the Belgorod region.



But happiness did not come on this. We had meetings with several agricultural producers, with a local agricultural university. We went to people and talked about our project, that we are making a robot that monitors plants in greenhouses, and everyone told us that this is some kind of sucks, and no one needs it. Therefore, we gradually stopped interacting with them, we did not have any implementation. The people who participated in this gradually dispersed. But we won the competition, and the expert jury gave us good advice - this helped us to believe in the very idea of ​​the product. Gradually, as we worked on this and talked to the industry experts with whom we came in contact, we realized that the idea of ​​a robot is really not very good. An agricultural robot is very expensive to develop, it is very difficult to do so,so as not to interfere with the production process. And few people are willing to pay big bucks for such technology.



But we already had ready-made developments related to the definition of plant diseases from images. We decided to make a simpler technology, which ultimately turned out to be in demand and made Fermata what it is now. This is the solution - when we install in greenhouses or in "indoor farming" (this is when real estate is converted into greenhouses) sensors for collecting accurate climate data and cameras for visual control of plants. Such stationary cameras are much cheaper than any robots, but at the same time the information that is collected from them is of sufficient quality for accurate monitoring of plants in production.



The main function of our product is to identify diseases in plants, but, in addition, we solve some of the problems that farmers face during production. We are trying to replace the entire decision-making process, which is based on visual monitoring, and make it automated. So, sometimes farmers need to determine the flowering phase, or control a special form of plants. All this can be done using the analysis of data from cameras. Sometimes the climate helps us in this: sometimes it affects the form or some diseases. We can also integrate visual and climate data to predict yield and guide farmers on how to behave to reduce costs and increase profitability.



This technology, which we came to after abandoning the robot concept, has already received support from investors, and we are developing it further. Another problem that we encountered quite quickly was associated with the Russian market: it turned out that this is not the best platform for piloting agricultural machinery, at least the one we do - aimed at reducing costs by reducing crop losses and reducing the number of people involved in routine work. This can be effective if the harvest is very expensive, or people are very expensive, and the decrease in the number of people employed in production (even by 1-2) is noticeable. There is no such thing on the Russian market. Many plant varieties, which are the most common, are not very expensive, so the savings due to our technology are insignificant for farmers.Labor is also not very expensive. Thus, the value of installing our technology on the territory of Russia is not very high. Therefore, initially, the negotiations that we conducted with the farmers while staying at home were not very positive. People did not really perceive us and did not see any benefit for themselves, because it really would be low.



But when we started trying to talk with farmers outside of Russia, we realized that we had a really cool solution. This was facilitated by the fact that I was in Israel. It was also a shake-up: to leave the cozy Moscow community of startups, where everyone is friendly to each other, it never occurs to anyone to ask you for money for advice, everyone calmly introduces each other to their investors.



The situation was completely different when we got to the startup events in Israel. We went to the first agrotech event together with my colleague who helped me with business development in Israel; came to this event, and it looked as if we were someone's children who had no one to leave at home with. Mostly there were adult men, 50+, who have worked in agricultural companies for many years, have vast experience, understand perfectly well what a conditional Bayer or Syngenta needs, and make a highly focused startup for a specific corporation in order to sell it there later.



When we found ourselves among them, we realized that our idea of ​​competition in the startup market was not true. There are a lot of very cool, professional people. And this influenced our further decisions regarding what to do with the team; we realized that it is imperative to involve experts, especially in the area of ​​finance and business, where it’s not technology and erudition that decide, but experience, knowledge, contacts. We also realized that it is very important that people with international experience are involved in this part of the team if you are interested in the international market. Unfortunately, it is very difficult to find people capable of providing a high level of international contacts.



While working with the Russian market, we immediately tried to start looking for people with whom we could pilot the technology - so that we would have real contact with manufacturers, so that we receive feedback on what they really need, and not come up with their requests ourselves ... We found a very friendly to us, at that time, company - "City-Farmer", which made vertical greenhouses for the production of herbs and mushrooms.



We made a free pilot with them; they opened their site for us, where we learned a lot - we understood how production works, what tasks are facing agronomists. We put together a large dataset there and figured out how to do it.



We have a dedicated person on the team who is only responsible for data. When we are working with any new project, when we are talking about some new type of plants or a new region - because, naturally, everything is different from region to region, and the tasks are also different - this special person does the research and finds everything he can find in open data.



Therefore, when we come to a client, we already have at least a prototype of a working solution. After that, we conduct intensive data collection, together with the client and his agronomists, and mark up the data. We have a special solution for data markup, so that the whole process is high-quality, intensive and fast. The process of collecting tasks from the manufacturer, processing the relevant data, building models that will solve the tasks facing them is very important.



Accordingly, when we started working, we immediately tried to find ourselves a commercial partner, on the basis of which we could develop technologies that would be needed and useful. Then we began to look at the decisions that we make - in which markets, in which regions they can also be useful. So we came up with a ridiculous, as it seemed to us, idea with medical cannabis: this is a rather marginal market, here the cost of the final product is high. Therefore, if you reduce production losses even a little, it plays a big role. It is useful for the farmer, and we can earn more, since our business model is based on the fact that we share with the farmer the profit that he additionally receives thanks to our technology.



So we realized that cannabis is a promising direction. We found another "pilot" in Israel, also free, where we worked out all the technology. After that, we, being already more intelligent than before, overgrown with a network of agronomists, with whom we now work, and which tells us what problems need to be solved, what features need to be added to the technology at each next stage. So we gradually made a solution for cannabis, which we are now actively promoting - this is a very attractive direction for us, from a commercial point of view. There is medical cannabis both in Israel and in Europe - projects where it is grown for cosmetic purposes, and there are the same important tasks.



If we talk about more classical plants, such as greens or vegetables, then we have to look at what is the cost in a particular region of these products, and what is the cost of labor there. Therefore, now we are focused on the European market - we understand that people are expensive there, and the effect of using our technology can be high.



One of the challenges facing farmers is to control the presentation of the plants. It is necessary not only to ensure that there are no diseases: the plants must look perfect. Here, our technology is also effective. If we talk about greens, then there are completely specialized producers who are developing plants for expensive restaurants, and it is very important that all plants look perfectly beautiful, so that the whole salad is even, green, beautiful.



Or, for example, there is an interesting problem in mushrooms: the mushrooms we buy are often quite ugly, and because of this they are cheaper. If a person builds a premium production, then it is important for him that all hats are dense, beautiful, of the same shape. We also solve such problems - we monitor the type of plants, even if it is not associated with diseases.



Why greenhouses and not fields? It seems to me that the task with fields is more complicated. The main question here is how exactly to set up cameras to keep track of the plants. In the fields, the main way of solving this problem is drones, but it seems to me that the hardware part is not quite ready to make a profitable offer to the farmer. To prevent drones from being discharged, not lost, they were not blown away, and at the same time they did not cost crazy money. Therefore, while we are working with greenhouses and do not focus on the fields.



Regarding R&D tasks: we are very interested in the direction associated with analyzing the composition of plants. We are conducting research in this direction, this is a rather important task for a number of plants, including those that are medical, but not only. For example, for wheat, it is important to know what its protein composition is. It is important for farmers to know in advance what is happening with the composition of plants in order to control the climate, to select the conditions so that the final cost of the product is as high as possible. Therefore, one of the most important R&D tasks that we see before us is the development of technologies that will allow us to do this in real time, to predict what is happening inside the plants based on the analysis of data from different spectra. Plus the analysis of environmental data and the integration of this data with each other in order toto predict and advise farmers on what to do to make their product as expensive as possible.



The tasks that remain with the person are a very important topic of ethical nature for all companies that are engaged in AI and labor automation. They are sometimes asked questions: Doesn't it seem terrible to you that you are developing technologies that replace people? But it seems to me that we are doing the opposite thing: we are making technology that allows people not to do boring, uncreative work that does not require the involvement of intellectual resources.



It seems to me that a person will never be caught up by machines in creativity, flexibility, the ability to create something that did not exist. Therefore, it is very cool to do things that allow people to do creative work, and not waste time, life, strength, health on what machines can do instead of them.



Q: What about automation technology for animals?



We thought about it. Early on, we thought about both animals and fish - in fact, there are many interesting tasks in fish. But now we are trying to focus as much as possible. We don't even spray ourselves between many different plants; we now have two main focuses - greens and cannabis. We consider and lead development in the direction of vegetables only because we have the potential for development with a major strategist, although working with classic vegetables is a very difficult engineering task. It is very important for us that we succeed - but in order to succeed, we need to limit our wishes. Therefore, we have chosen what seems to be the most commercially viable; defined the type of plants, defined the markets we go to and try to stick to them.



As far as I know, there are already many existing solutions in animal husbandry, very cool, related to AI. They hang sensors on animals, monitor their health, determine fertility by cameras, by gait. Very interesting tasks. Pisces seems to me a potentially commercially interesting direction, but I'm not an expert in this. So far we are limited to the priority group of plants.



Q: do you work with all types of cultivation?



Not with everyone. We do not work with open fields, because we do not yet understand what method of installing cameras to control plants in the fields will be effective. Before the epidemic, we had a relationship with one large American manufacturer of field equipment, but it all stalled. We are now focusing on what happens indoors - greenhouses and indoor.



Q: what did the startup live on before the first client? What were the first orders?



The startup still mostly lives on investment funds. We expect to reach profitability in a few months, but now the main thing is investment money.



The first commercial stories began after a reorientation from the Russian market to the western one, which happened not so long ago. We now have two fairly large commercial projects. There is also an important point here: we do not work directly with farmers, we work with technology manufacturers and greenhouse manufacturers.



If we were to install the equipment ourselves, we would have to have a large staff of people; As we focus on a large number of markets and do not want to limit ourselves, it is very important for us that the installation process of our technology is fast, efficient, and that someone else is involved. We want to stay in the software development area. Therefore, we work with companies that can provide the installation of the equipment necessary for our technology to work. Accordingly, these are greenhouse manufacturers, because our technology can become part of the greenhouse, or companies that make technologies for the care and control of plants. We talk with them about it like this: you make light, or heating the roots, or another cool technology that allows you to grow plants better, better, and we make technology,which allows collecting feedback from these plants. That is, we can find out how they feel, and we can tell you - and the end farmer - whether everything is going right, or some processes need to be optimized.



Q: what are your current competitors?



Complex issue. There are several companies in the world that claim to make similar technologies. When we first had our European team, we had this idea that they brought: we said that we were farmers and called different companies. We tried to understand, communicating with them, what they are really doing, how serious our competition is, what money they charge for their products - this is the most important question. Now it seems to us that the competition is not very high. There are many separate companies doing something similar; But as far as we understand, even those companies that say they use AI have a low level of machine learning in their projects. Therefore, we feel confident in general, but do not relax and try to move quickly and efficiently.



Agriculture is generally a fairly conservative, non-digitalized industry. It was very interesting to observe the dynamics of its change during the two years that we have been working; when we just started talking with big companies - even in the West - we were faced with skepticism, a lack of understanding of why this is necessary. But now, when we come to a corporation, we often hear that people there are just thinking about how to come to an agreement with a startup. And we are very lucky that we already have developments, results, metrics, models, and we do not just come and say that we want to do something. Now is a very good time for agrotech.



Q: what cameras do you use? Tell us more about the infrastructure



We have the following story with cameras: initially we did it all ourselves. But now we are trying to stay in the software area, because we have expertise in it - we believe that we are strong in data science, but not in installing and choosing cameras. There are very cool people who specialize in this; there is a Latvian company that deals with video surveillance - installs security systems in airports, in other large institutions, and they know everything about cameras. That is, how to install them, what problems may be, what to do if there is water, and so on. Now they are our main partner in the camera area. Together we choose cameras for our tasks; they provide the whole piece with selection, selection of cameras. This was one of the main correct decisionswhich we accepted - to stop doing this work within ourselves and instead attract high-level experts.



We also have climate sensors from the hard infrastructure. These include a standard set of parameters that you need to know in a greenhouse - temperature, humidity, CO2, lighting, and so on. We make wireless sensors that allow us to take information from the greenhouse in high resolution - we can scatter them around a lot and make three-dimensional maps of what is happening with the climate.



Until recently, we developed sensors ourselves, but then - especially after the partnership in the area of ​​cameras gave us a lot - we realized that we also did not want to do this part within the company. This is despite the fact that at the initial stage of work it seemed to us that this was almost a key element, and we even had patent protection for those sensors that we developed for our first clients. But then we realized that, in principle, we are not great experts in this, and activities related to hard are less clear and more complicated. Therefore, now we are organizing a partnership with another European company to produce this type of equipment for us.



Q: were there any problems with patents or patent trolls?



Not yet. Basically, the whole history of AI-related patenting is very complex; now we are throwing our efforts to solve this problem. Moreover, we have a lot of cool internal developments in terms of algorithms that are important to protect and which are the key value of the company.



Q: how did you validate the idea in the market and seek feedback?



We were just looking. I can't say anything smart here - it was painful and for a long time, especially at the beginning. They looked for friends at first, met with different people, told them, they told us - it sucks, met the next, they said the same. We tried to understand the logic in this - whether people simply do not like that we are telling them about something new and wasting their time, or there really are reasons why this cannot be profitable. Actually, we made the right decision - abandoned the robot - precisely because many people told us that it would not work.



We tried to talk to the market right away, tried to offer free pilots at the initial stage, we didn't talk about money at all. We tried to get at least some kind of expertise anywhere, because we didn't have an understanding of agriculture. We talked with universities. Now we continue to constantly validate ideas on the market - we have ideas about new features, and we communicate with clients about this.



We already have a network of people-consultants with whom we communicate; and there are directly those teams with which we interact. The business development process is about validating ideas, and this often requires cold contacts. There is nothing wrong with them, it really works - although when we were in the "incubator", it seemed that this was some kind of nonsense. Oddly enough, after hundreds of messages sent, you can really find a client or at least validate an idea if you have an interesting product.



I'll tell you about how all this is related to cancer diagnostics and biotech, and how it all works together. In fact, the technology and understanding of how to work with all sorts of data in agriculture initially, as I said, came from the idea of ​​working with data from cancer patients. Another project that I am a co-founder is called Smartnomica, and there we are developing technologies that will allow diagnosing cancer patients.



The idea is as follows. There are a large number of people - and not only cancer patients - who have gone through the path of standard medicine, and they have failed. Either they were not given an effective treatment, or the doctors could not understand what was wrong with them. These people need a slightly different approach; it is important to go beyond standard protocols and understand what can be given to these people by plunging into scientific research. It is known that scientific articles have been digitized since the 1980s, and there are over 40 million published articles in the medical field. They contain a huge amount of data that can be useful to each individual patient.



I joined a project that at that time was engaged in the treatment of difficult patients (like a medical clinic), as an AI specialist. Together with another co-founder, we had an idea: to automate the process of finding the necessary data, which is done by specialists - doctors, scientists - for difficult patients.



Make this process at least partially automated to help clinicians conduct research more quickly and efficiently for complex cases. Around this, we have built a company that we established in Riga in March. In fact, we only planned to think about establishing a company, but due to the epidemic, we were no longer able to leave Riga, and we had nowhere else to go except to develop the company. We now have a clinic in Latvia, where I am now: we connect doctors around the world with patients around the world to provide treatment. And, accordingly, we are developing technologies that allow our doctors to work more effectively with the data of complex patients, to carry out high-quality diagnostics and select effective treatment.



So what we do in agrotech is related to what we do in oncology. We also have many interesting projects here. The decisions we make for our patients sometimes turn into separate biotech startups, of which we now have a whole 1 (but there will be more). All of the technologies that we use for our patients are related to integrating different types of data - which, from a technological point of view, is similar to the story that we make at Fermata. There is visual data, genetic data, and scientific data that we need to combine to give the patient the most effective therapy possible.



Q: About Fermata: at what volume of production is it profitable?



It depends not so much on the area as on the type of plants and the region in which this occurs. The area is not so affected; although this may affect our desire to be involved in the project: we try to select large manufacturers in which we see great potential for scaling our technologies. The area of ​​a specific production is not so important: it may be small, but there may be many such production facilities.



Q: growing in soil or hydroponics?



We have experience with both methods, it is not that important.



Q: maybe it would be a good idea to work with city farmers? There people are younger, open-minded, loving innovations



This is true, although we do not work with the farmers themselves, but with those who produce the city farms that they use. This is one of those areas of work that we are carefully considering.



In particular, this is one of the reasons why we work well with medical cannabis: the people who are involved in these industries are often more technically savvy than usual, and their production is more technological than the production of classic vegetables, for example. Therefore, they take the idea of ​​introducing AI much easier.

Q: Hyperspectral cameras are usually used for your task - they are very expensive; What cameras are we talking about and how many are used per farm?



The question is complex and I cannot say how much is used per farm. This is always different, we look at the specific truss design. The spectrality of the cameras also depends on this. Multispectral cameras are more of a drone solution. Greenhouse solutions are not like a big market, there are no hard standards. We often use ordinary cameras for our projects that shoot the visible spectrum + IR - this is enough for most tasks. It's not that expensive.



The number of chambers depends on the design of the greenhouse and the type of plant. I said before that classic vegetables are an engineering challenge; for example, a tomato greenhouse is a jungle. It is very difficult to observe each individual plant and leaf, because the plants grow 4 meters. But the salad is small and flat, much less difficult. Just like strawberries - or the same cannabis, which can be effectively observed from above and get an idea of ​​what is happening with the plants. Usually, one of the cameras that we use - depending on the height at which the cameras can be placed - is enough for 200-400 square meters.



Q: unfortunately I have not found any patents for Fermata.

These are not patents yet, these are patent applications pending.



Q: do you have vacancies?

Yes, if anyone is interested, you can send applications to valeria.kogan@fermata.tech. We are looking for interesting and talented people all the time.



Q: not thinking about doing mycology?



I don't really understand the question. We also analyze problems of this kind visually. We analyze whether there are insects on the leaves, and whether there is a nutritional deficiency, and various infectious diseases - we monitor all this and are able to determine.



Q: are you planning a mass product like a cloud service?



This is a very good and very big question. The fact is that a cloud service for our technology is not the best solution. This was where we started. As it turned out, we collect a lot of data, and it is important for us to collect high quality images. Therefore, a clear cloud is not optimal. We have a distributed solution between cloud and on-site; on-site analyzes the images and sends the results to the cloud. Now it seems to us the most optimal approach. But this does not make the product less popular.



Q: mycology is about mushrooms [not parasites]



We did one project on mushrooms - I even mentioned that mushrooms have interesting problems. Now we do not have a massive mushroom project, but we have experience and interest in development in this direction.



Q: is it possible to determine the needs of plants before the appearance of pathologies?



I think this is not the task in our current approach. We try to automate the work that people do. That is, they tell us that they want to see leaves that look like this, and receive notifications about it. And our task is to do it more efficiently and quickly than specialists do it manually. That is, we are not solving the problem of predicting the state of plants. It's only when we use visual data in conjunction with climate that this task is sometimes solved. But more this concerns some kind of nutritional problems. With regard to pests or infectious diseases - we determine this only when it has already happened.



Q: tell us about Fermata and the dataset: what do you collect, how did you mark it up? And about the neural network - is there a collab, to see to roughly understand what you are doing?



At first we collected the dataset from everything that is open on the Internet. For example, from all the contests that were held on kegel and on other resources, in particular, Asian ones, there are many interesting things. We collected pictures from all sorts of agronomic forums, from different encyclopedias - we tried to collect everything, mark it out.



When we got clients, after some time we worked out the procedure for working with them. For example, a client has a need to detect some very rare malicious piece that can kill the entire greenhouse, and there are not many images of this piece. And they themselves, as soon as they see it, take pictures and send us images. Also, our specialists regularly travel - especially if we are working with a new plant - and take photos and train the staff inside, so that we are constantly sent data to replenish the dataset and improve the quality.



First of all, we have internal specialists for marking. We have a network of agronomists - specialists in different technologies, different types of plants and their problems in different regions (this is important, because the same plant can have different problems in different countries and on different continents). We also actively involve our clients' agronomists for marking. They are ready to share data, because they understand that it will be better for them: if an agronomist spends 2 hours a week to tell us what tasks they face, then we will be able to better solve them.



About the collab - probably not now. But the question is interesting, I did not think about it. I will consult with the guys, maybe something like that should be organized.



Q: what exactly is the development process and the result?



The result, the end product, is the system. We come to the greenhouse, install cameras-sensors, and then the client has a dashboard that shows everything that happens to the climate and plant health. Alerts come there with information about when and how. The client can mark and track what he is doing with these plants. That is, a product is an environment in which an agronomist works, and in which all the state and health of plants is visible.



Q: you said that one camera is 200-400 squares, but is there any experience with vertical multi-tiered farms?



Yes, there is such an experience (in fact, we started with this experience). Of course other cameras are used. In conventional greenhouses, it is important to have a large zoom and cover a large area, while in vertical greenhouses, a different approach. They use cheaper cameras, which allow us to observe the plants in different rows, realizing that we will not be able to install the camera very far.



Q : what do you think of microalgae?



Unfortunately, I am not an expert in this, I cannot answer.



Q: Do you see, as a result, the need for automation of soil moisture monitoring, automation of dosed irrigation?



Yes, this is a very important task. We see such companies (especially working in Israel - the center of such technologies) primarily as our partners. They are engaged in the automation of irrigation, and we monitor the plants and can give them feedback on how well their systems are working. It seems to me that there are many successful companies that solve this problem, but I cannot say anything more (not an expert).



Q: which state is quantitatively now, where are you geographically located, is there (or is it planned) your own hardware R&D department?



The staff of the company is about 13 people now. Geographically, this is a complex issue, we are located in many places. There is a development team in Moscow, there is a team in Israel, there are people in Germany. Sometimes we involve outsourcing development to solve more standard tasks outside the team, in order to be focused on machine learning ourselves. That is why teams are so widely scattered around the world.



Hardware R&D is unlikely now. I would like to focus on software, this is due to where we are now moving and where we want to be, and those whom we see as our potentially most interesting partners are just companies that make very cool hard, and are very strong in this area. We don't want to create unnecessary competition and waste our time exploring new areas instead of perfecting the part where we are experts.



Q: my question was just about vertical farms and how many were used there.



I can't say the exact numbers. If you have a practical interest, you can write off later.



Q: have you looked at business with northern agricultural producers (greenhouse complexes in the Urals and further north)?



We tried to talk with several northern manufacturers, but at that time they had no success. This was at the beginning of our journey. We did not return to this story, because now we are focused on those types of plants that are our priority.



Q: how about wheat, sunflower? Drones?



I am skeptical about this. We tried talking to different companies that make drones and tried to find a solution for that. It is the approaches with drones that seem risky to me now, and we are not going in this direction. But there are some interesting ideas about how this kind of monitoring could be done in the fields, but now this is not a priority for our R&D.



Q: what education do you need to have to work in your field of artificial intelligence (which programming language, which area of ​​mathematics)?



Our programming is mainly in Python. It seems to me that any area of ​​mathematics will be good at programming, there is no hard requirement. In principle, the people we are now looking for for an internal team to work on AI - machine learning specialists - should be able to read the article and implement what is written in it. That is, not just using standard libraries and standard existing networks. It is important for us to be at the forefront of what is now in technology.



If you have additional questions, ideas for cooperation, or if you want to join our team - write (valeria.kogan@fermata.tech), we will be glad to meet you.






All Articles