Send AI revolutionizes data extraction from documents with artificial intelligence

In today’s corporate world,extracting accurate data from complex documents has become a necessity. However, many companies struggle with challenges such as accuracy, security and handling unstructured data. In response to this need, a Dutch startup called Send AI has developed a customizable platform that allows companies to train artificial intelligence models toextract data from specific documents. This platform recently received backing from Google’s Gradient Ventures, opening up new opportunities for Send AI in the document processing sector.

The challenge of unstructured documents

In the highly regulated insurance industry, for example, companies often have to process a variety of document formats, from PDFs to paper files to photos taken with smartphones. This unstructured data is difficult for even humans to process, but a fully AI-based approach can lead to errors in handling claims or complaints. Pre-packaged document processing software is often designed for more common documents found in different industries, making it unsuitable for certain use cases. This is where Send AI comes in.

Send AI’s customizable solution.

Send AI offers a customizable approach to extracting data from documents, allowing companies to train computer vision and language models to recognize specific documents and extract and validate relevant data. If necessary, humans can be involved at each stage through a web interface to control and review the process.

According to Thom Trentelman, founder and CEO of Send AI, “Validation can be as simple as checking whether an expected number is really a number, or it can be more sophisticated processing of a registration number in a database to see if there is a match. Any insecurities will be flagged for human review.”

How Send AI works

Send AI offers cloud-based software that companies can use through APIs to email documents. Once received, documents are visually enhanced by Send AI before being sent to language models for classification and data extraction.

The main target of Send AI are large companies that have a large amount of documents to process. However, any company facing a high volume of documents can benefit from this technology.

Advantages of Send AI

Send AI offers several advantages over other document processing tools on the market. While large language models such as those offered by OpenAI can work well for subjective tasks such as summarizing or answering questions, when it comes to accurate processing of large volumes of documents, these models can prove unreliable, slow and expensive.

According to Trentelman, “with Send AI, we leave it up to the customer to build their own solution.” Send AI is based on smaller, open source models that are trained by the client itself through manual processing of a small set of documents. This approach results in powerful and accurate models that can be adapted to the client’s specific needs.

In addition, Send AI ensures that its clients are in control of their own data. While other LLM (Large Language Model) solutions might mix training data from different clients into a single model, Send AI uses isolated models for each client. This means that Send AI can pinpoint exactly where the data resides and delete it on demand, ensuring the privacy of sensitive data.

Pricing and development plan

Send AI offers its clients a basic credit-based plan, in which clients pay for each processing step. This allows them to differentiate between processing a 50-page PDF and processing a single snippet of text. According to Trentelman, “Our models are cheap, fast and reliable, so we can implement them on a customized basis for each client. This gives clients control over their data and performance, which is why we perform well in regulated industries such as health insurance and government.”

At the moment, Send AI is in private beta, but it has already landed major clients, including insurance giant Axa. With a team of seven, the company plans to double its staff later this year before officially launching the product.