@entropybender
pretty much exactly a year ago (aug 2024), concept for future.news
i thought the hardest and most novel part was going to be creating markers on the chart that reflect significant news/tweets/signals
however, like always, the most difficult was transforming the data and getting everything to be performant in a time series db. especially updating aggregate data such as volume, moving averages, etc
the benefit of having clean and unified data, and only a few tables/views to reference was that the LLM agent we built was able to generate proper SQL queries to answer the long tail of questions efficiently