We use to search for web search (google) and document search (within an enterprise). In web search, we have an experience where we get precise information in Answer Cards and simple language that we understand and the list of links to click through. However, when it comes to enterpise, we are still far away from Google-like experience. In this case, we aim to improve search experience for enterprise products through Answer Card and Semantic Search techniques.
We wanted to design a way to allow our early users to experience the power of Semantic Search. One way to go about it is through directly integrating semantic search in any product. Another way is to build a standalone experiment that offers the semantic search experience. We had a couple of constraints around data and infrastructure because we decided to go the latter route of building a standalone experiment.
The team managed to present this experiment to a larger audience inviting immense interest from product teams for deeper integration with their respective product. At present, a couple of collaborations are in progress.
Find out the riskiest assumptions around infrastructure and data needs for your experiment at the earliest. You may end out scaling down the offerings in your experience depending upon those constraints. Also, find a way to crowdsource the feedback collection part to train your ML model. The more feedback points you have, the better the chances of ML model prediction.