At OXOS, the challenge of navigating a sea of documentation is a familiar one for every team, especially the software team. Brand documents, guidelines, and industry regulations are to be considered in every step. Ryan McArdle, a Machine Learning Engineer at OXOS, sheds light on how he’s transforming the approach to documentation management by implementing AI. When questions are answered faster and with increasing reliability, time is saved and operations are optimized.
The Need for Innovation
“We have so much documentation that we need to understand to do our jobs well, from design specifications to FDA regulations. Just keeping track of where all of this information is, let alone what it says, can be a full-time job,” Ryan explains.
Introducing RAG and LLM
To tackle this challenge, OXOS is leveraging advanced AI techniques. Ryan shares, “We are creating a Retrieval Augmented Generation (RAG) chatbot to use the power of a large language model (LLM) to manage the wealth of information contained in our documentation.”
How It Works
Ryan elaborates on the system’s functionality: “The retriever finds the relevant documents that a user needs information from, and the language model/chatbot uses these documents to inform a response to the user’s questions.”
Benefits for OXOS
Ryan highlights the advantages for the team, “This helps us quickly sort through a sea of documents to find the information that we need when we need it, all with a natural language user interface.”
Current Implementation and Future Plans
While the system is currently focused on the software team, Ryan discusses the potential for expansion: “The goal is to get the full wealth of our documents available with this tool. For now, it’ll stay with just software for small-scale testing, ironing out intricacies before expanding.”
Ensuring Accuracy and Relevance
Ryan addresses the challenge of maintaining up-to-date and accurate information: “It’s really hard when you’re working in this industry to verify that everything is good to go – because accuracy is an absolute must.” The system’s ability to prioritize and understand revisions is crucial for ensuring reliability.
He explains this is what they’re looking into currently, “If you’re giving conflicting documents, say like revision A through G, and you’re not giving it a way of understanding that revision G is more important than revision A – that will be an issue. What we’re testing right now is having the system recognize the newer, prioritized versions.”
Next Steps
Ryan outlines the current status: “We have a small framework in place, but we’re continuously testing and refining based on user feedback. Building something robust starts with small-scale testing.”
A Glimpse into the Future
He concludes with a glimpse into the future: “We’re also developing a knowledge graph to handle structured data, enhancing our capabilities even further.”
Conclusion
OXOS’s AI implementation is an innovation win for tackling complex problems. Employees make a difference daily in moving us forward along with technological advancements – and this is just one example. We’re excited to see how this project develops as it expands to different teams across the company.