benchambers
Re:Extraction Matters Most
on Wednesday February 28, 2024 @04:47PM (#39543880)
Ran some experiments to see how different things affected the quality of results from a retrieval-augmented generation (RAG) pipeline. Post shows results and details / code for how to run similar experiments yourself.
Re:Always On: Four Pillars of Real-Time Generative AI
on Tuesday November 07, 2023 @12:32PM (#38179923)
Author here. When I started hearing about Generative AI, I was really excited about the potential for "always on" agents eagerly assisting me rather than passively waiting for the next question. I wrote this blog-post as I've been thinking about what it will take to build such agents, that watch everything happening in real-time and proactively help.
Curious what others are thinking in this space. Have you been thinking about using AI in a real-time manner? What kinds of proactive agents would you like to see?
Always On: Four Pillars of Real-Time Generative AI
Re:Timelines: Declarative, Temporal Queries
on Tuesday May 30, 2023 @02:38PM (#36128729)
In this post, Therapon and I dive into a bunch of temporal queries showcasing some really powerful concepts -- using multiple event streams in a single query, creating windows delimited by properties in the data, and performing temporal joins.
Re:Timelines: An Evolution in Stream Processing Abstraction
on Tuesday May 09, 2023 @07:29PM (#35881706)
Hey -- it's Ben (one of the authors of the post).
The most interesting thing about timelines to me is how naturally they work at multiple layers. This post shows how well they can be used for understanding values over time. But we'll see that the same requirements we discuss -- that data is ordered by time and grouped by entity -- make them a natural fit for expressing queries and implementing efficient, incremental execution.
We'll be around -- if anyone has questions or thoughts please share!
Re:Show HN: Kaskada – modern, open-source event processing
on Thursday March 30, 2023 @04:00PM (#35378101)
Interesting that you mention DSLs like Drools and Esper. One of the things we'd like to do is explore ways to bring the complex/composite event-processing (CEP) functionality to Kaskada. SQL does include "match recognize" at this point, but it is still pretty unwieldy to use.
With the shift towards general event-processing, we'd really like Kaskada to make it easy to express queries like "how many times do X happen within an hour after Y with these properties" -- the kinds of things that Drools and Esper provided!
Show HN: Kaskada – modern, open-source event processing
https://github.com/kaskada-ai/kaskada
We recently open-sourced Kaskada – an event-processing engine built on Rust and Apache Arrow. It features a high-level, declarative query language designed specifically for reasoning about events in bulk and in real time.Unlike SQL, an aggregation over events produces a timeline indicating the value at each point in time over the events that have occurred up to that point in time. The query...
Re:Ask HN: Who is hiring? (October 2019)
on Tuesday October 01, 2019 @03:33PM (#21129597)
KASKADA | SEATTLE | FULL-TIME | ONSITE
Looking for: Full-time engineers for full-stack web-application and data-processing backend
Kaskada is a Seattle-based startup revolutionizing enterprise machine learning through the use of real-time data. Our team is delivering an end-to-end machine learning platform powering feature engineering and productionization. We are hiring for mid-level and senior engineers for our frontend and backend teams.
As a member of a small startup, you will define and implement a new, interactive experience enabling data scientists to design, visualize, and collaborate on new features for their machine learning models.
Re:Ask HN: Who is hiring? (September 2019)
on Wednesday September 04, 2019 @08:05PM (#20882646)
KASKADA | SEATTLE | FULL-TIME | ONSITE
Looking for: Full-time engineers for full-stack web-application and data-processing backend
Kaskada is a Seattle-based startup revolutionizing enterprise machine learning through the use of real-time data. Our team is delivering an end-to-end machine learning platform powering feature engineering and productionization. We are hiring for mid-level and senior engineers for our frontend and backend teams.
As a member of a small startup, you will define and implement a new, interactive experience enabling data scientists to design, visualize, and collaborate on new features for their machine learning models.