Overview of View

View is a platform that accelerates your AI journey by integrating your private enterprise data assets with a comprehensive processing platform and built-in extensible AI experiences.

View is deployed on-premises in your data center, in your colocation facility, or in your private cloud instances, and allows you to connect your private data with AI without needing to send your data beyond your security perimeter.

With View, you can:

• Quickly ingest, process, and prepare your data for AI consumption
• Chat with your data to yield insights
• Build AI-powered experiences to improve customer satisfaction, employee efficiency, and more
• Rapidly deploy internal conversational AI experiences using your data
• Understand and discover your data assets

How does it work?

View is comprised of a series of microservices, largely within four categories.

• Ingestion: Before data can be used with an AI experience, it must first be ingested and then processed. Ingestion within View can happen by uploading data into an S3 bucket, via our data crawlers, or programmatically through our API.
• Processing: Data ingested by View is processed through a series of steps to extract meaningful metadata, relationships, features, and to generate vector embeddings.
• Persistence: The resultant data assets from data ingested and processed by View are stored in a variety of repositories for later use, including relational, metadata, graph, and vector.
• Interactions: Built within View is the ability to interact with processed data through a large language model (LLM) powered natural language chat experience. Alternatively, you can build your own and integrate using our APIs.

Details of each are covered in the following sections.

Ingestion

Ingestion is the process by which data is made accessible and consumable by the View platform. View currently supports three mechanisms for ingestion.

Need a different interface? Let us know!

• S3-Compatible API: View includes an S3-compatible storage service, allowing you to define buckets and upload or otherwise manage objects. These buckets can be assigned rules that dictate how objects stored within are processed. View's S3 interface is compatible with the broader S3 ecosystem including tools like the AWS CLI, Cyberduck, S3 Browser, and more.
• Data Crawlers: View provides crawlers to connect with your data where it resides, and currently supports local filesystem, Network File System (NFS), Common Internet File System (CIFS), S3, and Azure BLOB repositories. With View's crawlers, you can schedule crawl operations and specify filters to dictate what gets ingested and how it is processed.
• RESTful API: View is an API-first platform, and data can be ingested in real-time by using our RESTful APIs. These APIs can be integrated directly through your shell scripts or within your own applications using our available software development kits (SDKs) which make integration easy.

Processing

Processing encompasses a series of steps that marshal ingested data into the forms necessary for consumption by AI. These include:

Need a different processing step? Let us know!

• Type detection: The foundation for any processing is built on top of a confident understanding of the type of data with which you are processing. View uses magic signature analysis and other techniques to examine content to understand its actual type.
• Semantic cell extraction: Regions from the source data containing coherent and related information are extracted into what are known as semantic cells (for instance, a paragraph in a document, or, a PowerPoint slide). These cells are then chunked into reasonably-sized pieces to balance processing throughput and fidelity of results.
• Metadata generation: View generates metadata, called Universal Data Representation (UDR), for every piece of ingested content. This metadata contains a variety of features of the original content, including key terms, inverted index, schema, flattened representation, geometry, and more.
• Embeddings generation: Vector embeddings are generated over each semantic chunk using a large language model (LLM) assigned to the ingestion task. This process converts source data into vector space, that is, floating point numbers with both a magnitude and direction, which is foundational to how AI interacts with data.

Persistence

Persistence is the act of writing data derived from processing to the appropriate repositories for later consumption. View automatically persists data to a variety of repositories in the form of metadata, relationships, and vectors.

Need a different repository for persistence? Let us know!

• Metadata: View includes a data catalog and semantic search platform called Lexi which holds metadata and features for all ingested content. This information is then searchable using many dimensions as input including key terms, schema constraints, key-value constraints, and date ranges.
• Relationships: By default, View includes a LiteGraph graph database to store information about each ingested object, including data relationships, meatadata and features, and repository information.
• Vectors: View includes an instance of pgvector to persist vectors from processed documents and other useful metadata.

Having the trifecta of data (e.g. metadata, relationships, and vector) allows View to provide a best-in-class retrieval augmented generation (RAG) pipeline to ensure the highest levels of performance, scalability, accuracy, and fidelity.

Interactions

Interactions with data are the primary driver behind any AI project. With View, four primary means of interacting with your data are provided:

• View Assistant: Built into the View dashboard, users can select knowledge bases with which they'd like to have a conversational experience, allowing them to ask questions of their data to yield useful insights, summaries, and predictions.
• Bespoke using View APIs: Homegrown applications can expose AI-powered interactions fueled by source and processed data through View's comprehensive API platform, which includes a robust API for retrieval augmented generation
• Standalone, Packaged Experiences: View provides a template by which conversational experiences against a given knowledgebase, along with accompanying configuration parameters, can be rapidly deployed, allowing your users to interact with knowledgebases without providing them direct access to your View deployment.