Introducing Search AI by Base64.ai, an innovative document search engine that finds all relevant documents based on natural language queries, using either words or your microphone. Search AI surpasses the limitations of traditional document search engines, such as Google Drive’s search tool, which relies only on the file name and OCR from a document. What makes Base64.ai Search AI unique is its ability to search based on casual queries, such as “show me driver licenses that aren’t expired”. Base64.ai Search AI understands documents beyond text including the document type (e.g. “search driver licenses”), key-value pairs (e.g., first name=John, last name=Doe), custom taxonomy, data normalization, and AI features (e.g., number of signatures, number of faces, image blur and glare).
A Groundbreaking New Way To Search Documents
Search AI is a smart document search engine designed to help customers find the right documents in a Flow using specific queries, making it easy to quickly retrieve any document within a Flow. Since our customers often deal with thousands of documents, manually sorting through them for specific info can be a big challenge. To tackle this, Base64.ai has simplified the document retrieval process, letting users find what they need with a simple query.
Base64.ai understands document information beyond mere text
Popular search engines and document management companies, like Google or Dropbox, have many limitations in searching through documents.
For instance, if a user saves a New York Learner Permit to a Google Drive folder and attempts to search for “Driver License”, the document will not appear. This limitation stems from Google Drive primarily relying on text-based search, where the file name or text within the document must exactly match the search prompt.
OCR-based search will not find a New York Learner Permit as a Driver’s License because it doesn’t contain the exact match.
Base64.ai considers document classification and understands the information within the document, ensuring that all relevant documents are retrieved in response to a search. Base64.ai search classifies New York Learner Permit under New York Driver’s License, thus it can be also searched as a US driver’s license or driver’s license document categories.
Another powerful feature of Base64.ai Search AI is searching the document metadata that isn’t immediately available in OCR. Tables, Faces, Signatures, and other properties of the document are only visible via Base64.ai Document AI, which can extract the features and use them for searches like “Show me NDAs that are not signed”
How does it work?
The process involves four main steps.
- Document processing (A-1): Base64.ai Document AI processes documents and returns the results in a standard manner so they can be organized for search
- Indexing (A-2): Elastic Search is used to organize document data alongside other permission information such as Flow ID (so that only you can search in Flows you have read access)
- Query (B-1 and B-2): Converts manual search parameters and human sentences to JSON queries
- Search (B-3): Executes the search
How to use Search AI
There are multiple ways to create a query. Use speech-to-text to search by simply speaking into the microphone or type a sentence manually. Users also have the flexibility of employing direct filtering to specify document details based on type and criteria, tailoring the search to their specific needs.
In the following depiction of Search AI, the question “Show me all expired driver licenses” was typed into the search bar, where it was then indexed for the most relevant information. Once the information was confirmed, Base64.ai retrieved only the documents that are expired.
LLM is used to convert human sentences to a query format. In our example, this would correspond to “{“documentType”:”driver_license”, “expirationDate”: {“lte”: “2023-12-06”}}”. LLMs can convert the document type, standardize fields such as expiration date, and place today’s date in yyyy-mm-dd, so it can be sorted for date.
The user can further change the search query manually or add new search vectors. When choosing model types, users can specify the document processing model(s) for the search. The Criteria filter plays a key role, in determining how the prompt relates to different fields. Users can input both common and specific fields, connecting them to specific values using operations like <= or !=. This comprehensive filter covers various elements such as OCR text, numbers, faces, and signatures within a document. The order of criteria offers flexibility, allowing users to arrange fields in descending or ascending orders for effective organization of search results.
The search can be extended to other parameters, such as OCR length, page count, file mime type, number of tables, number of faces, number of signatures, image blur, image glare, etc. The capabilities are only available in Base64.ai Search AI as other document management systems cannot see the document in depth without AI.
Enabling Search Function
To activate Search AI, navigate to your Flow, go to Settings > Edit This Flow > Basic Information, then scroll down to the Document search engine section and simply click the checkbox. Before returning to your Flow list, make sure to update your flow settings to successfully apply the changes.
Getting started today
To learn more about Search AI, visit our Flow page. Feel free to contact us for more information or navigate to the Signup Page to create your free account to get 100 free pages to test our AI.
About Base64.ai
Base64.ai is the leading artificial intelligence platform to automate document processes. It understands all document types worldwide, including IDs, passports, invoices, checks, and forms for banking, insurance, logistics, travel, gig economy, and RPA companies. In seconds, the Base64.ai platform ingests documents, classifies them by type, extracts information using OCR, Generative AI, Large Language Models, computer vision, and deep learning, verifies the results, and integrates the data into customers’ systems. To learn more, visit www.base64.ai.
