Eni works with AWS to develop its collaborative geoscience platform

Right. Welcome everybody. Thank you so much for being here. My name is Hussein Shall. I'm a Chief Technologist, Head of Subsurface Transformation at EWS Energy. I'm here to talk to you about what we're doing with E&I around energy transformation.

I'm very excited about the talk today. We're here with a distinguished speaker from E&I who has a wealth of information and experience around energy transformation. Apologies if the lights are too bright to look everyone in the face, but I'm pretty excited to be here.

First, I want to give you a run through of the agenda. We're going to talk about the changing landscape of energy. Then we'll invite Francesca to the stage to talk about E&I, what we're doing around geoscience, the project we're taking on together, and the challenges we're hoping to solve.

I'll go through the architecture, a deep dive into the services we've used to make this possible. Then we'll finish with our aspirations and next steps for this collaboration.

The modern energy landscape is undergoing a profound and rapid transformation. The importance of energy security, reliability, accessibility, and affordability is driving digital transformation and technology as an enabler. Coupled with the urgency around sustainability and energy transition, there is critical demand for clean energy sources, propelling organizations like E&I to balance economic considerations with environmental responsibility.

This intricate balancing act requires companies to innovate swiftly, leverage technologies, and form strategic partnerships to meet the demands of a future where secure, reliable and sustainable energy is an imperative. This is driving significant digital transformation for energy customers and partners like E&I to drive lower emissions, lower costs, improved production, and minimized risks.

The integration of advanced technologies like ML, AI, HPC, and IoT signals significant strides towards sustainable energy solutions driven by technology advancements.

E&I is an Italian multinational company pioneering global exploration and production in challenging environments. They are recognized as pioneers in diversifying their energy portfolio around solar, wind, etc.

I'd like to welcome Francesca Salvi, Head of Geological and Geophysical Studies at E&I, to talk about our collaboration.

Francesca: Thank you for this opportunity. I'm a geologist at heart. E&I covers the entire energy value chain from exploration to end products and services in 62 countries with over 50,000 employees. This came from our founder in the 1950s to explore and discover new resources for Italy. The same vision based on people, technology and sustainability is our mission today.

E&I is targeting carbon neutrality by 2050. We are committed to an energy transition that is socially fair, preserves the environment, and provides energy to all who need it. These pillars of sustainability, security and affordability are the energy trilemma.

What role can geosciences play in this complex framework? Geoscientists have a unique skillset to take on the uncertainties of earth systems and are crucial to implementing the sustainable development goals.

With current technologies, we can acquire large quantities of data. Through this data, we can apply a holistic approach to define cleaner energy mixes to support the transition. Geoscientists can identify risks, characterize geothermal energy, identify aquifers, and more. But how can we approach this challenge?

Today's technologies allow us to acquire huge amounts of data for real-time monitoring of the earth's surface and subsurface. By observing, learning and exploring from the data, we can obtain a global vision - the real challenge of the future.

Can geological workflows be digital? Seismic data is digital native, but geologists prefer tangible, 3D environments like rock outcrops. How can we combine this analog attitude with innovation?

Geological workflows are heterogeneous - they span scales of time and space. Reconstructing the past from present evidence makes it hard to standardize. Workflows use broad, disjointed data connected by metadata. The data is heterogeneous so we face resolution challenges. Consistency is key to accelerate projects.

This complexity makes geology ripe for digital innovation to improve efficiency, strengthen creativity, and connect people and data. Today's disjoint tools limit this. A collaborative ecosystem could improve our approach - a connected, open data platform leveraging 3D, ML/AI, and more.

Geico is a collaborative platform for geoscientists to support sustainable use of resources and mitigate geo-hazards, with applications from exploration to training. It will make structured and unstructured data accessible to minimize uncertainties and will be open and extensible.

Most importantly, it will put humans at the center - providing information tailored to users' jobs and drawing from diverse sources. This stimulates creativity while reducing bias towards existing processes. The platform is designed so intuition shines through.

"But in this case, the target and the aim of the ecosystem is also to highlight something that is unknown to challenge the assumptions of the user and to highlight the outlier in order to support and stimulate the creation and the innovation.

So the platform, it will be also uh the technical solution that will improve uh all the activity of our company and i at global scale worldwide. But also it will be also the desired solution for sharing geosciences data between share uh the stakeholders in a safe and secure environment.

Thanks also to the capability of AWS and supports and the platform will consider international standards in order to be open and extensible and uh synergies. And the collaboration between ENI and AWS has provided an excellent opportunity for growth and an excellent example of complementarity and synergy uh on different aspects.

Such as, for example, as we, as we have said, the capability to work world wide at global scale. And again, also the from one side, uh ENI, the uh experience and the knowledge in geosciences domain plus the experience in the creation of collaborative data management systems with the capability of AWS and the guarantee uh um came from AWS of multilevel grant administration and security.

So it's a space and uh also of course, uh sharing the mutual experience in innovative technologies involving artificial intelligence and natural language programs.

But now let's see some uh preliminary capsules of the model that we are developing uh uh focus of go focusing on unstructured data.

These uh uh we can see the landing page, for example, uh from which the users can start to dive and interrogate the systems based on uh a customized profile and uh querying some natural language questions.

For example, right now, the system supports two type of data sources, documents and propriety databases containing geological information and in specific domain following the natural language questions.

In this case, uh in this example, we see uh a natural language question focused on a quantitative data. For example, which is the porosity of the fine rocks. The page of results is populated proposing the relevant paragraphs from the internal document.

In this case, this is the reason that some words are great in yellow. You can see highlighted the keywords included in the questions and on the left, you can also as user select and prefer the type of data that are visualized and also uh make other preference in the searching mode.

And um another way in which the system by leveraging the generative AI can support the research. And the study of the geoscientists is a proposed summary uh recapping. And um the main uh in in this case values of the policy that i'm working for with the minimum and maximum values that are reported in the in the available documents.

Each document is related with the document metadata. So there is always the possibility to check the sources of the information that are reported. And uh each document is also related to the so-called business entity or business uh object annotations that depends uh uh could be uh wealth formations, basins and so on.

Again, another example uh that highlighted the capability of the system to propose you even more longer ques uh summaries depending on the question that we are asking in that case again, a natural language question, quantitative focusing on the quantitative values.

In this case, uh the system um was able to split in three items, the question and so the answer. But uh the user again, if uh is not satisfied of the, of the, of the sum or the the answer can uh uh reask a new generation of summary changing, for example, some input.

And since we have seen the geological, uh the geographical location of course is an important factor in our job. Uh the system is also able to propose uh the um the documents, the relevant relevant documents answering to the re the the questions again, natural language question uh with the proper geolocation in since if in the document is in is reported was included at least one geographical information.

All the objects are geoloc so proposing paragraph, relevant paragraph but even images if needed. And um another question, another example just to highlight that the system is able not only to answer natural language questions focused on the quantitative but also on the qualitative information.

For example, in this case, the question was uh related to the main orientation, some geological features. And so again, the system proposes you some of the documents that are reporting this information in that specific geolocation with a summary of the main trend and the main information including in the documents.

The answer are also um it is possible also to have some answer uh through the images instead of the text. Indeed, following a natural language questions, the system can propose you some of the images that are classified within the documents between seismic.

For example, that is one of objects of interest for us uh maps, geological maps or plot. And um again, in that case, the system is able also in this case to propose you a summary of uh the main information included in the images.

Uh and from the captures, we decided also to integrate the system with two tools, additional tools able also to extract some quantitative information from the plot. For example, in this case, that is uh uh uh plot digitize. Uh thanks to this tool, the user can um focus on a specific plot.

And after a very, very quick uh um set up of the axis extractor the value xy values in a separate table. And uh similarly, we can also focus on some tables inside the document and that we can see on the right and um extract from the table, the values of the header uh inside the table if needed.

Uh the user can also edit the values if it's not uh correct because of some uh uh for example, in the in the extraction. And these two tools can absolutely support the user to recover all the historical data data set that we have usually in the company in form of pdf in papers in order to be able also to reconstruct for example, the history of one specific asset and so on and integrate the old quantitative information in new modern analysis analysis.

Now i leave again the floor to Hussein in order to see a little bit better, which is uh which architecture is uh behind of all these systems, please.

All right. Thank you so much Francisco. So we wanted to give you just kind of a start with the high level block diagram of the different entities or different system modules within what you saw so far. Um and so if you want to start with, at the front end, you've, you've seen the um um AAA UI that's designed specifically for a natural language processing type of interrogation supported by a context uh definition.

And so in order for us to provide context to that search, we leverage Amazon Textract to extract the text as well as the tables. So there was very specific training to recognize the table structures in those documents. Um but we are also using Amazon Comprehend to identify the business entities needed um uh like well, names and basins and other entities to enrich and provide a more context based answers.

Um additionally, for the image classification, we're using uh recognition and a very specific custom made image classifier uh through Amazon Recognition service uh that allows users to search based on images based on titles based on things uh associated with that particular image and what what i want to you know, say it's essential to note is that we're trying to build this on a serverless uh architecture that allows the scalability and the security of ingesting and, and, and feeding into it, you know, hundreds of thousands and millions of, of, of documents at the end of the day and pulling data from uh proprietary systems as well as or systems of record beyond unstructured data.

Um to go a little bit into the specifics and the architecture design. Um you know, um we, you know, and this is really encompassing both front end and back end capabilities um and building on on the services that i mentioned like text track com, brand recognition.

I do want to highlight Kendra uh which was very essential in enriching the uh the summaries and the um the the the more natural language interrogation uh of the process. And so what we're doing is we're leveraging uh Kendra as a natural language processing engine and we're also incorporating an open source um uh large language model with uh llama two to enhance the uh the summary and the demonstration uh that Francesca showed uh on each one of the search results.

Um and also as you can imagine, you can see the kind of uh usual suspects of API gateways to handle API calls and management. Uh SageMaker JumpStart is used to enable the integration with the llama model um and further and further.

Now our oh sorry, did not realize these were supposed to be. Uh yeah. So again, text track, comprehend recognition each was building for a specific uh purpose and then really integrating the generative AI extension to existing services. And this was all done again, this is early stage uh in the in the project uh prior to bedrock and what you heard today or this week uh with uh Amazon Q which would replace some of that, right?

And so we're leveraging in particularly with Kendra and, and the language model is the um retrieval augmented uh uh generation or RAG architecture to combine the two and create that synergy between them. And now we can start exploring Q and some of the other services to make it a little bit easier, less custom built.

Um but our, you know, job doesn't stop here. This is really the beginning of the partnership and the relationship and this collaboration. Um our aspirations is to start integrating this with third party applications from a data source and potentially from a workflow standpoint.

Um we also want to start incorporating, like I said, some of the more managed services uh generative AI services uh like bedrock and Q et cetera. Um we're also um wanting to incorporate and, and make go potentially a more of a canvas for collaboration than just a, a search engine as it is today. Uh that's the first step and then we take it further.

Uh so, pulling into a kind of creating it as a canvas of collaboration that pulls different data and rates uh data based on experience of previous geoscientists and, and users and create a collaborative space that pulls that data becomes the driven part of it.

Um and lastly, which is really, really important is the integration with um open standards and industry standards like open subsurface data universe. Very specific uh oil and gas standard for those who don't know ah because we want to ensure the interoperability and ah adherence to industry frameworks and best practices and API s ah to enable that interoperability that i talked about with the third party applications.

Um so I want to hand back the mic to Francesca to go over what we're doing next and uh some of the key benefits for um uh for G thank you.

So, so let's two slides just to conclude uh which are the key benefits that we uh that we foresee in this tool and the use of this tool.

First of all, uh mm the use of the collaborative ecosystem in our opinion will improve the uh our risking capabilities and process uh since better quality accuracy, completeness and correctness of all the information through an integrated multi scale management data and workflows could improve the quality of our uh geo model and uh our, our results of our activities and projects.

And at the same time, we can minimize the cognitive bias uh of the user that we have seen and sometimes could in who and leads to the same wave of uh working and uh results obtaining results.

Second, we improve our efficiencies since the optimization of the search times and the information evaluation with the support of gen AI technologies. Uh it, it will be uh pragmatic and effective uh improvement of our efficiency of working saving time.

And uh thanks uh third, thanks to the transversal of the system. As we have seen, uh it will be applicable and usable to the different business line within the companies starting from the more conventional and traditional lingus to the carbon capture storage, for example, or the development of the renewable energies.

But uh also um outside the industries, as we have said, uh uh with uh forcing potential adoption outside within international and uh uh organization and institutions.

And as the last point, uh as we have seen, uh uh we are often requested to accelerate the deployment and deve development of our discoveries and the, and the project. And this uh request uh the uh the, the full data consistency of our data in order to guarantee the high quality.

And uh also the mm the right use of the data within the long process that could be also interactive with different resolution between the different technical functions that are working on the same asset.

So what's next? Uh as uh Husain already said that we are just at the beginning of the journey. So JECO is uh uh we are developing, developing JO uh and the ecosystem is rapidly growing, expanding the technical capabilities and progressively including different type of data.

We will add uh of course uh structured data. Following the finalization of the unstructured data, we will improve our capability of analysis. And uh we will also uh we have already included in the system, the tool to visualize what is the virtual outcrop model that is an acquisition, a digital acquisition of the uh works on the surface, the outcrops and we will improve also a tool in order to extract and quantify all the measurement on the virtual outcrop in order to the risk and better constrain our geo model in the subsurface.

Again, we will include uh different the course the laboratory results in order to expand also the range from the outcrop to the microscope data and uh expand also the data providers of the structure and structural data.

And finally, with the wells at the end of the day, all the phases and the package will contribute to an effective ecosystem of insights heterogenic, populated and to propose an unbiased space to create our futures.

So this is the end of our talk and presentations. Thank you so much for uh the attention even it is late in the in the evening. And if you have any questions, please feel free to ask. Thank you."