The Company

Veritone (NASDAQ: VERI) is a provider of artificial intelligence (AI) technology and solutions. It is perhaps best known for its media and entertainment applications used by companies such as ESPN, CNN, HBO, NFL Network, the San Francisco Giants baseball team, and many more. Veritone processes over 100,000 hours of video each day for its customers.

The company also has clients in the government, legal compliance, and advertising markets. The company’s big idea from the start was to provide access to hundreds of cognitive engines through one common software platform and to “democratize” AI by making it available to any organization. This evolved into aiWARE, a cloud ecosystem of hundreds of cognitive engines that the company calls an AI operating system. The value proposition for aiWARE is to manage all the complexity, cost, and deployment of all AI models for you. Veritone was quick to point out it is not trying to compete with AI giants Microsoft, Google, or Amazon, but that it instead provides one platform that can integrate and manage those companies’ AI models alongside those from Veritone and other third-party providers. As vendor models often change, a unified platform should offer users more flexibility.

The company is expanding its aiWARE business unit and launching into cognitive capture use cases for intelligent process automation and adjacent markets. aiWARE includes content extraction and analytics models for vision, speech, audio, text, biometrics, and data. Audio and video content is a hot growth area for information capture, and Veritone enters it far ahead of the traditional document capture vendors.

The Technology

aiWARE is a cognitive model development and deployment platform that is optimized for unstructured data processing and is accessible from both AWS and Azure commercial and government clouds, with an on-premises deployability option on the horizon. Veritone calls it the “OS for AI” ecosystem, with hundreds of ready-to-deploy AI engines. aiWARE provides a single development interface for all AI models from Veritone and other leading AI engine vendors. Designed for business analysts and “citizen” developers to create low-code workflows, the platform is also scalable to handle massive amounts of video, audio, images, text, and data in near real-time. The platform consists of six layers (see Figure 1):

1. Data ingestion layer. Real-time adapters capture and ingest unstructured and structured data stored on all major data storage platforms, plus YouTube, social media, live broadcasts, and news media platforms.
   
2. Cognition layer. This is the heart and soul of aiWARE. The platform enables access to hundreds of cognitive engines spanning 25 different cognitive capabilities, comprised of over 40 Veritone proprietary models, third-party models provided by large and small software companies, and proprietary engines from Veritone customers. By setting cognitive engine standards across all models – what Veritone calls their VTN standard – aiWARE simplifies the development, onboarding, and deployment across all cognitive capabilities and focuses on performance. Onboarding proprietary engines for private use has been simplified with the Veritone Developer application.
   
   Here is a list of the cognitive engines most relevant for intelligent document processing:
   
   • Text cognitive engines: content classification, entity extraction, keyword extraction, unstructured text extraction, sentiment analysis, summarization.
   • Vision cognitive engines: image OCR text recognition, object detection, logo/brand detection.
   • Speech cognitive engines: transcription of audio or video files in 70 languages, speaker recognition, speaker detection, speaker separation.
     
3. Orchestration layer. This is where we found more of the Veritone secret sauce. The Veritone Conductor benchmarks the available cognitive engines to determine the optimal combination for a given data set. Then it applies proprietary machine learning (ML) algorithms to intelligently route data through the most effective workflows of cognitive engines and produce a single normalized output.
   
   Why is this so important? Ensemble learning (also called multi-engine interference learning) is typically used to improve the accuracy of an AI model’s output. With ensemble learning, every cognitive engine within a cognitive capability must be run to provide the lift in accuracy. As you can imagine, this is highly inefficient and could easily make costs skyrocket. This is not viable in an enterprise production environment.
   
   Veritone uses conducted learning, an approach that takes into account not only accuracy but also efficiency and cost. With conducted learning, costs are optimized by predicting the minimum set of engines required to achieve the best result.
   
4. Metadata search and indexing layer. aiWARE creates a data lake that offers search by cognitive capability and preserves data insights into AI engine outputs with time-correlated cognitive metadata indexing. It can trigger custom workflows off prescribed events between aiWARE and third-party systems through an event-driven framework.
   
5. Integration layer. This layer consists of the Automate Studio AI workflow tool and APIs. With the Automate Studio visual programming tool, business analysts can tap into the APIs with little to no coding. aiWARE APIs access and extend the intelligent capabilities of aiWARE with exposed GraphQL APIs. They can upload data, summon cognitive engines, and access cognitive metadata.
   
6. Application layer. aiWARE natively supports three application categories:
   • a. AI-enabled interaction analytics. These solutions for conversational intelligence built on top of the aiWARE platform produce automated insights around customer interactions (topics, keywords, intent) across common communication channels such as phone calls, video conferences, email, and social media. Three pre-configured solutions are available: conversational compliance, social media insight, and contact center insight. The customer, Veritone, or a Veritone partner can also build custom solutions.
   • b. Turnkey industry applications. There are currently over a dozen Veritone-developed turnkey business applications powered by aiWARE and designed to meet specific industry challenges without the need for AI expertise. These applications include Veritone Redact, Illuminate, IDentify, and Contact for their government, legal, and compliance business; Veritone Digital Media Hub, Attribute, Discovery, and MARVEL.ai for their media and entertainment business; and Veritone Forecaster, Optimizer, Controller, and Arbitrage for their energy business. Veritone will soon release Veritone Verify, a smart biometric SSO that combines multiple cognitive authentication methods simultaneously, including facial and voice recognition.
   • c. Extensibility solutions. These self-service tools enable enterprise developers and partners to leverage aiWARE’s configurable platform architecture and create solutions customized to their business challenges. The tools provide the flexibility to create AI workflows and integrate them into other applications, and to onboard and deploy new cognitive engines. Veritone will soon release an engine benchmarking tool on the aiWARE platform that will compare performance of engines on the platform, helping the user with AI model explainability.