Those of you actively using Baselight daily may have realised that last week, a new “AI Chat” button appeared in the upper navigation bar of the platform. Surprise! Baselight AI v1 is available in preview! We decided to get the feature out of alpha silently so all of our active users could be the first ones to try it before the bigger announcement coming next week.
I think at this point, how to use Baselight AI be self-explanatory for everyone, but let me share a few examples that you can start tinkering with to get you started:
- Here’s an analysis of all the insider insider buy and sells transactions of different AI-related stocks trying to answer the question of if we are in an AI bubble (and if Michael Burry is right shorting some of these stocks 🙂 ).
- I’ve been saying this a lot lately: with prediction markets “sport betting” has become more of “sport trading”. I have to admit that Baselight AI has been helping me a lot with analysing upcoming NBA games, and trying to find an edge when “trading” on NBA-related prediction markets. Here’s my analysis from last night.
- And why should you trust these analyses? You may be wondering. Well, there’s a few “goodies” under the hood that allows you to “audit” how the AI reached a conclusion:
- You can follow Baselight AI’s reasoning flow:
- And inspect the specific queries and data points used for the analysis
- Finally, you can easily inspect the generated charts throughout the conversations, and even save them into a dashboard that you can periodically update with most recent data:
In short, a breath of fresh air over the current state of LLMs where, I don’t know about you, but I see myself spending more and more time double-checking facts to prevent potential hallucinations. I am super excited with this release, and I really hope that you enjoy it as much as I am doing.
Any feedback, suggestions, or improvements, do not hesitate to let me know. And I would love to see some of the really cool analyses you are coming up with (use that share button on the top right of the screen, and share your chats with all of your loved ones –and ideally me 🙂 ).
Improving catalog searches
As we are onboarding more data daily into the platform (over 57k datasets already), search is becoming more of a priority to us.
We not only want to surface the best datasets through our platform search, but we also want Baselight AI (and external LLMs connected through the MCP server) to surface the best datasets for their analysis in a single call. We’ve been seeing that Baselight AI needs a few searches in order to land the right dataset, which increases the time from question to analysis.
To improve this, we are already testing in development some nice improvements to our search engine through:
- The improvement of semantic search through the use of embeddings.
- An improved ranking logic to prioritise high-quality datasets over lower-quality ones.
We are staging the release of these improvements in the coming days, so you should start seeing how finding the best dataset becomes easier for you and your agents.
A new landing page and more data
Finally, if you go to https://baselight.ai you’ll see the beautiful landing page that I alluded to in my last update. And this is not its final form, we are releasing a few more cool sections with our product offering and use cases in the coming days, so stay tuned and do not hesitate to make it a visit every now and then.
And I couldn’t end one of these letters without a glimpse of the data that is coming to the platform soon. Next we are getting: Kalshi markets, e-sports (mainly league of legends for now), and improvements to our EVM blockchains, decoded tables and SEC data. Do not hesitate to thank Paulo personally for the amazing job he is doing seeding Baselight with the best structured data we can have 🫶.
Alfonso de la Rocha – CTO
https://x.com/adlrocha