.png)
4 mins.
About 15 minutes into one of our Lettria demo calls, the conversation always sounds the same:
“I get why your tech is better, I see how it can help us, but I have to sell it internally. We’ve been burned before, so I need an implementation plan that really works to convince our CEO to get on board.”
Over the past 5 years, I’ve lost count of the number of conversations had with clients and potential clients about AI PoC failures in their companies. It’s been especially bad over the past year or so, as the marketing and claims around LLMs like ChatGPT have far outpaced the technology’s effectiveness in enterprise applications.
Ensuring client satisfaction thanks to our technology’s applied usefulness is why Lettria isn’t currently offered as a turnkey SaaS. What do we do instead? We start by nailing one of your key use cases with our Pilot Program.
Creating your first verifiable, trustworthy GraphRAG solution
As soon as a potential client contacts us for a demo, we start the conversation about their current problems that could be solved thanks to GraphRAG. We don’t pretend that we’ll immediately solve every problem, or that their team can simply plug Lettria into their existing stack and have the whole team using it tomorrow.
Instead, our goal is to find one big problem where Lettria can:
- Help develop an effective database, using the unique documents and data the company needs to query;
- Become an easy-to-understand AI solution that team members use confidently;
- Have a significant impact in terms of time and resources saved.
Generally, a pilot program takes about 12 weeks, from the initial outlining of the project through development and deployment. It generally works as follows 👇
Weeks 1-4
We begin by defining the project needs, detailing the current problem, who it affects, and the outcomes that will constitute success. With this done, we work with the client to collect all of the data sources that they wish to include in the completed interface.
Once that data is collected, our NLP technology can begin the hard work of properly parsing it, ensuring that unstructured data becomes structured in order for it to be correctly interpreted by a machine. This gives us the basis for an enriched ontology and the initial knowledge graph, using the client’s data as well as Lettria’s existing resources.
With structured data and the initial ontology and graph in place, we can move onto testing the client's advanced, custom GraphRAG solution.
Weeks 5-8
The first prototype begins to be tested by both our client and Lettria’s team of experts, including data scientists, lexicographers, technical directors, and project managers. This happens in a series of feedback loops, as questions are posed and answers are evaluated according to a matrix (correct, partially correct, partially incorrect, incorrect). Each session lets us note various improvements that then go into the next iteration.
Throughout, the goal is to continue enriching the ontology and improving relationship graphs in order to confidently deploy our GraphRAG technology for a given use case. This second phase brings us to a solution that is ready for final testing and deployment.
Weeks 9-12
The final iteration is completed. Lettria’s engineers work to produce the appropriate API documentation, as well as to complete the interface within the client’s information system. We assist with the rollout and make sure users are comfortable using the new tool.
The end result is a verifiable, trustworthy AI application running within your company. There’s no black box results, no hallucinations, no misreading of charts and data. Query results come back quickly, and are displayed alongside their knowledge graph and the context from which they were found.
One focused success for a team, one giant leap for the company
Of course, we don’t want things to stop there, as Lettria can be used to solve a wide variety of problems in any given business.
A successful pilot is just the first step. In working with our current clients we’ve seen how it can serve as both the launchpad for improved brainstorming on where else Lettria’s technology could potentially be useful within the company and the step-by-step blueprint for successfully bringing GraphRAG to bear on other problems.
Importantly, even as it will get easier to apply Lettria to more use cases within the business, once the pilot program is completed we don’t just send our clients off on their own. Our team of experts keeps working hand-in-hand with them to both develop, test and refine new applications – all part of our “humans-in-the-loop” conviction that the future of AI is augmenting humans, not replacing them.
