Jun 19, 2024

Successful AI Orchestration - Contextual’s Building Blocks

Successful AI Orchestration - Contextual’s Building Blocks

Expanding on our blog post on why AI Orchestration is critical, this post digs in a bit deeper into how Contextual solves for each of the required building blocks of a successful enterprise AI solution.

Federated Data

AI operates best with a range of data inputs, including both enterprise data and third-party sources. Simply connecting to that data is insufficient—the platform itself must be designed for data transformation, extraction, standardization, and classification as assumed predecessor steps for running effective AI solutions. Controlled data inputs are just as critical as controlled prompts and model responses.

Contextual’s approach leverages a dynamic ‘Object Types’ data structure. This JSON-schema-based strategy ensures that even if external data formats change—or if the response you get from AI has additional new, or unexpected data elements—information is not lost, and can easily change your data structures on the fly to match evolving patterns, including by using our integrated AI Assistant to just describe the changes you need. Have a new input from a third-party system that you’ve never seen before? Changing the format of a specific data point because you swapped CRMs? Never a problem.

AI Copilot inside the Contextual schema builder

Federated AI Models

Production AI solutions in the enterprise often require a sequence of interactions across multiple models or tools, including general GPTs, hyper-targeted or functionally tuned LLMs, RAG-enhanced assistants, and machine learning models.

Contextual’s connections framework allows you to dynamically swap between different (but potentially similar) AI solutions in order to compare—even in an A/B framework—the performance of distinct LLMs. Our ability to send workloads to distinct agents in series or in parallel means you can assemble a staged set of AI processing tasks or functions that satisfy your ultimate AI solution objectives.

EXAMPLE: This workflow in Contextual uses WebPilot AI to browse and summarize a client domain based on lead form and then uses RapidAPI Classify LLM to classify the business based on Regulatory Environment, Country, and Industry.

Flow editor inside Contextual

Asynchronous Event Processing

Effective AI solutions often occur ‘out of band,’ with some models taking significant time to process and return their results. This variance requires the ability to direct actions asynchronously as events (inputs and outputs) occur. Its critical that an underlying message bus that supports this so that AI processing can seamlessly occur alongside existing enterprise systems be in place.

Contextual has removed all of the complexity of leveraging an Apache Pulsar event messaging infrastructure while still delivering all the benefits of high volume, high scale, and high throughout processing. Through object triggers or simple send-to-agent event nodes in the Contextual workflow engine, AI solution developers can drop events onto topics that have guaranteed processing, retry logic, and controls around order of operations.

Micro-Service Workflows

Given the industry's pace of change, decoupling individual steps into microservices (data transformation distinct from model processing) is critical. These steps can then be dynamically swapped between—or run on—multiple models with monitoring for accuracy and drift. This gives enterprises the greatest flexibility and avoids lock-in. Just because you are a Microsoft shop does not mean Anthropic isn’t a better answer to your specific need.

Contextual’s low-code, visual-based editing capability, called ‘flows,’ encourages the development of hyper-specific and easy-to-understand individual microservices that can be assembled into a broader solution. These flows can also be easily updated and maintained by any team member.

Distributed Delivery Endpoints

Not all AI is chatbots. Enterprise users need the results of an AI processing workflow delivered alongside their existing work patterns. If AI assists in pricing recommendations, the results will sit inside the CRM CPQ function. If AI prioritizes work orders, the directions and notifications sit inside the work order management platform. Making it easy to get the targeted and explicit results back into the existing enterprise systems is critical. 

From Contextual’s built-in tenant API to HTTP endpoints serving complex experiences with sophisticated styles, Contextual allows AI-assisted business solutions to present their results seamlessly into the existing enterprise system, tools, or workflows. This ability to run complex AI ‘alongside’ existing capabilities means the AI experience appears—in context—when and where users need it.

By delivering on this complete set of AI Orchestration needs, Contextual provides the fastest time-to-value for an enterprise or its supporting development teams to create powerful, proven, and ready-to-deploy AI solutions in weeks, not months. If you or your team are looking for an advantage in leveraging AI, Contextual is your answer. Reach out to learn more.