Train AI Models

Recent innovations in AI technologies for Large Language Models, Natural Language Processing and Machine Learning have made it possible to train models on business content. Trained models are tuned and will deliver precise, expert-level guidance that power recommendation engines, search algorithms and provide guidance to drive customer decisions at time of purchase.

There are technical challenges inherent in training AI models that can be overcome with automation. But a fundamental problem that many businesses face when integrating to public AI engines is that of keeping business data secure and private. As Large Language Models (such as ChatGPT) have demonstrated, any usage of those public systems inherently renders a company's private data insecure. That data is aggregated into the collective corpus of the public AI engine and is no longer under the control of the company that produced it. This is a very dangerous frontier for businesses who leverage their content as a competitive advantage within their industries. It cedes control of the content to those public AI engines and provides them with free acccess to crisp and insightful, well-curated content at the expense of the business using them.

Instead, companies would like to maintain control of their content to keep it private, secure and highly curated without ceding advantages to competitors or public, third-party aggregates. Private implementations of AI models would made it possible for businesses to not only own their data but also own the AI models and endpoints, allowing companies to securely administer access and usage of those endpoints for private use. Private AI models build on top of trained base configurations and layer private, curated data on top to fine-tune the model to provide expertise on a company's brand, products, messaging and content.

These private AI models can then be used for both internal and external objectives. Internally, the AI models can be utilized to support AI services that automate and enhance the curation and delivery of content for a company's editorial team. And externally, they can be used to support customer needs and decision/guidance services, such as generative search, recommendations, relational queries and similarity searches.

• Privacy of Data

  Public AI models aggregate your private business data into their collective corpus, gaining the advantages of your data and intellectual property while also forcing businesses to cede authority and control over the content.

• Ownership of Trained Models

  Privately owned, trained models would expertise and precision on a company's products, brand and messaging based on fine-tuning and test/validation using the company's curated content.

• Automation

  Automated bots and curation tools can leverage private models to assist with content generation, workflow and management steps to make your editorial team more efficient.

• Customer Engagement

  Smart AI-driven endpoints can assist customer-facing APIs to provide needs analysis, recommendations, similarity searches and decision support for web sites, applications, voice-activated interfaces and external APIs.

The Gitana Platform provides out-of-the-box facilities to instantiate and train private AI models that protect your data and operate entirely under your own control. Your private data is secure and never moves into the outside world. Instead, it is used to train and validate AI models that then provide downstream support to your own customer-facing services and APIs that back your applications, web sites and other integration points.

The following product features accommodate the needs for this use case:

• Private AI Models

  Own your trained AI models. Retain your content, training data and generated corpus to glean and improve, turning your private AI models into a powerful competitive advantage.

• Private Data

  Never cede ownership of your data. Your investment into highly curated, sharp and rich information is retained by your business as a competitive advantage. Tune your own model and surface decision making logic as you see fit (for your own applications or for internal use by your editorial team).

• Automated Sidekicks

  Instantiate one or more automated sidekicks to augment your editorial team and assist with content generation, workflow processes and editorial changes. Configure each sidekick with a personality and expertise, guided either by available public corpuses or AI services, or using your own private AI models. Chat with sidekicks to glean insight on your documents.

• Procedures

  Define scriptable processes that sidekicks can execute to perform very complex work tasks, such as writing new documentation, connecting nodes together in a relational hierarchy, applying taxonomy tags or metadata labels, publishing content or running test/validation against QA/integration environments. Let Sidekicks perform the grunt work so that your editorial team can focus on curation and mastery of your base of content.

The Gitana Platform provides an ideal foundation for training, validating and utilizing powerful AI models that are privately owned, secure and kept up-to-date with changes to your branch-based content store.