Let me share with you my story about finding the right AI Solution to improve our support self-service. The starting point of my journey was pretty common, many will read this and think “Yes! I suffer the same pain”.
At JFrog we produce tons of relevant and clever documentation. It’s all out there, available. And yet, we struggle to get to the right piece of information when looking for something specific. So this is when I realized that we need to bring in THE one to rule them all – aka a clever, learning indexing and search solution that would understand the intent of the requester.
Sounds like an easy task, right? Well it’s not. I quickly found myself looking for a needle in a haystack. So many solutions, some from simple indexing to other super complex, at the verge of esotericism, “we can help you predict the future” ones – all looking promising and offering advanced, top-notch features and functionalities.
Fast forward to 11 solutions considered, I had to face the reality: the AI domain was still very young and no solution in the market could exactly match JFrog’s use case – applying machine learning to indexed documentation with complex and fuzzy technical phrases and terms, which often includes rich logs and software errors.
Therefore 10 months ago, we decided to develop our own tailored in-house AI solution.
Applying machine learning to content
Artificial Intelligence (AI) is about understanding the intent of the user, with Natural Language Processing (NLP). Sometimes you can ask the same question in many different ways. For example, a search for “SSO Login”, should include results around “SAML”. We wanted a solution that could understand a query that would result in making a correlation between questions and the relevant content resources.
The magic of data science tools
Our AI solution is based on Apache Solr Search, which is a Lucene based search engine that utilizes the BM-25 ranking algorithm. Unlike older searching solutions that are more commonly dependent on the TF-IDF algorithm (Term-frequency / Inverse-document-frequency), BM-25 uses real Machine Learning principles. It is particularly successful at understanding the user’s intent by using Natural Language Processing (NLP), enabling a self-learning system.
All JFrog data sources being created by our developers, support engineers, customers, product, marketing, users and more, are now indexed into the new system, bringing the most accurate answer for your questions.
Snapshot of some of the indexed content in page numbers
*represents October 2020 and still growing by the hour
The JFrog AI solution is constantly learning. The more it is being used, the smarter it gets. See for yourself! Log in to the New Support Portal we’ve designed for you and enjoy the customized JFrog Self Service experience 🙂
Your feedback is at the core of our considerations, please email us at firstname.lastname@example.org with your thoughts and any suggestions on how we can further improve.
Want to contribute to making our content better? Here’s your chance!
JFrog encourages its users to contribute their knowledge entries to our content resources. This provides the opportunity of sharing and learning from everyone’s experiences. Got a great idea that would like to share? Submit it to email@example.com and we’ll review and get back to you on possibly making it public to everyone.