Case Study: Building an Enterprise-Grade Generative AI Platform

I took a moment to interview an engineer on our team about the work we recently launched for a globally distributed enterprise organization. While a lot of AI talk in Enterprise has been about process and tool adoption, others have found opportunities to build AI tools specific for them. Here's the breakdown of the exciting work our team is executing in the context of a platform we just launched!

‍The Challenge:
‍Like many enterprise organizations, our client wanted to operationalize generative AI for internal teams. Similar to organizations wanting their search tools to “work like Google,” if you’re reading this, you’ve probably been in a conversation or thought to yourself that it would be great to have a “ChatGPT, but for our proprietary information.”  Like search, a simple ask is not that simple to execute, and the benchmarks set on the user experience are the result of billions of dollars.

Not only did our client need to deliver a solution that avoids risking data exposure, governance issues, or system instability… they needed a solution that could: 

• Ingest internal documents
• Generate consistent, structured guidance
• Reference source materials for accountability
• Run entirely within private infrastructure
• Scale without degrading performance

Public AI tools were not an option due to legal, security, and compliance requirements.  On top of that, even some Enterprise License tools lacked the UX needed for adoption.  UX is what matters to your users, and when you’re delivering for a large operational userbase, the stakes are high to meet or exceed the billion-dollar benchmarks they have become accustomed to.

‍The Solution:
‍Clique designed and implemented a custom generative AI platform integrated directly into the client’s internal ecosystem.  We did this by starting with a prototype that allowed us to evaluate technical performance, work through fundamental architecture with manageable subsets of data. 

However, a key to the success in our implementation was conducting UX workshops with prototyped UI in parallel to evaluate user experience. In years of executing UI and technical work at the enterprise level, we’ve learned that while you can prioritize one over the other, both the technical architecture and the user experience have to deliver.  Our expertise in working in environments that are highly regulated with strict governance have allowed us to leverage methodologies we’ve developed to deliver rapid results that prioritize both.  The key was launching iteratively and involving users throughout the process.  While our engineers used iterations to validate scalability and performance, users became part of the experience development.  This avoids the “Grand Opening” you see in a lot of rollouts that build excitement in the short term for a launch, at the expense of long-term gain and adoption.  Users already know how it works, participated in the decisions that led to how it works, and organizations can focus on change management and continued iteration in place of rollbacks and fixes.

‍Key components of the platform we delivered included:

• A secure document ingestion workflow for PDFs, presentations, and structured inputs
  
  • This approach not only mitigated the need to undergo a huge document conversion process, it also ensured that no assets were omitted from being used in the platform.  
• A standardized prompt framework aligned to internal business “focus areas”
  
  • This was key.  Users did not have to become prompt engineers to get consistent results.  By effectively increasing the ease of use we were able to maximize adoption across the enterprise.
• A private large language model hosted within the client’s cloud environment
  
  • Privacy was paramount to the initiative, and now results in technology that will continue to improve from our client’s proprietary data.  This is a competitive edge that will continue to grow in value for our client, vs. a SaaS product where their usage improves the product their competitors may also be using.
• A modern frontend UI for document upload, progress tracking, and results consumption
  
  • By working with users, we learned about their surrounding processes and incorporated features into the UI to make their jobs easier beyond the generative functionality.  
• Asynchronous AI processing using a queue-based architecture
  
  • This was one of the larger challenges to overcome.  We deployed an architecture that mitigates performance issues due to spikes in usage.
• Live feedback loops and real-time status updates for long-running tasks
  
  • We uncovered an opportunity for the interface to be informative vs. purely transactional.
• Unified enterprise authentication ensuring access across disparate cloud environments
  
  • This enabled secure deployment for global users to easily login and use the platform. This also minimized adding new tasks for IT to manage.

The Result

The platform enabled internal teams to:

• Analyze past materials and performance data
• Generate structured recommendations for future initiatives
• Maintain data privacy and governance
• Trust AI outputs through transparent sourcing
• Use AI at scale without impacting platform stability

What began as a prototype became a production-ready enterprise rollout.  I should reiterate it’s serving a global user base and we went from prototype-to-production in less than 6 months.

At Clique, we’ve been the team behind exciting Enterprise UI and innovative technical builds in heavily regulated industries for over a decade.  Bringing that experience has allowed us to introduce and build AI tools responsibly.  As a result, integrated AI solutions for the enterprise has become an area of expertise at Clique. Our right-sized team structure allows us to move fast while leveraging our experience in secure platform architecture to deliver solutions faster and safer than massive teams.

Is there an AI conversation happening at your organization, an idea to explore or a problem to solve?  Contact us, I’ll bring folks from the team that delivered the above to our first call :-)  

If the opportunity is not a good fit for us, we’ll let you know, but will also share insights from everything we’re learning in real execution that can help you determine next steps!