Indra wants to revolutionize the Free Energy Market using AWS-based AI.

That's why Indra thought about developing a platform that would allow interested customers to upload their accounts and, based on an analysis, obtain automated cost estimates. “We began by trying to develop an in-house solution, but analyzing non-standard information became a challenge,” she explains, recalling that, in addition to each concessionaire having its own layout for the accounts, they were received in different formats, such as color image and B&W, text, .pdf, etc.

That considered, the Indra data team thought of a solution that applies generic data extraction from the different layouts employed and the use of generative artificial intelligence (AI) to make sense of this data, allowing its analysis and the automation of responses to customers. “Since the creation of Indra our processes have run on AWS, so we sought information with them to understand how they could help us with this project,” she reveals.

After the first conversations, AWS recommended Flexa Cloud for the development of the project, which aims to extract information from different energy bills, standardize them and provide answers to users. Ruan Medina, an Indra Energia data scientist, explains that the information extraction process was divided into two stages.

“In the first, we transform everything into text that can be read by the platform and, in the second, the data is identified and given semantics,” he explains, recalling that the process allows us to identify more than 180 different document models, all with the same data, but presented in a different way. “AI is fundamental for identifying this data and the semantics of this context,” he says.

The idea is to avoid that Indra has to manually collect this data for each new client. “When we realized the possibility of contacting Flexa, which was already working with Generative AI, and of inserting a tool into the process with the support of AWS, we found a form of technical development,” she celebrates.
Flexa Cloud CEO, Deivid Bitti, explains that two layers of artificial intelligence (AI) are being used to develop the Indra platform. The first, based on Amazon Textract extracts data from documents in different formats. “Here, we were able to create a process that extracts and converts the image into text and, in the end, returns a series of highly assertive data,” he explains, citing the examples of personal customer data indicated on the bill, energy concessionaire, consumption and contracted demand histories, invoice details such as fines and charges, tax rates, group marking and tariff framework, consumer unit identifiers, among others.
The second relies on Amazon Bedrock for the use of generative Artificial Intelligence (AI), a critical step in the process. Generative AI makes it possible to identify and facilitate the semantic processing of information extracted in various processes on the platform. “Bedrock was chosen because it allows us to quickly test a variety of foundational models and at a low cost. Bedrock allowed us to test different foundational models until we identified the best one,” explains Bitti.
The Anthropic’s Claude-2 foundational model, from was chosen after performance tests brought excellent precision results for Brazilian Portuguese. The 100K token context window was critical as it allowed us to work with all the entirety of data found in the energy bill, which can be several pages long.
The architecture also relies on Amazon S3 for secure and temporary storage for energy bills. Amazon Aurora in PostgreSQL stores all platform data in a structured way. “We are going to transform this system into an API where it will be possible to present the file and receive the data. It will be a 100% serverless solution, fully pay as you go, keeping up with increasing or decreasing demand,” he explains.  

Medina provides more detail and says that Indra will have modules to meet every step in the journey of its customers and partners. “What we have here is the customer research, identification and acquisition module, but we are still going to expand the frontend for these accesses, management of the benefits received by partners, and will also implement various connections to internal and external data to analyze the data extracted from energy bills. There is an entire development demand that will be used for this solution and we will have a lot to use from the AWS infrastructure in this process,” he predicts.