Co-Founder of https://teapawt.org/t.co/2HGY4U981l ✧ Professor https://teapawt.org/t.co/NYeeXKrz5m ✧ Space Computing https://teapawt.org/t.co/wf9QXN1Hix
Stockholm
Joined March 2008
- Tweets821
- Following330
- Followers874
- Likes3.3K
Pinned Tweet
We shouldn’t have to wait for the next frontier model to trust AI with real work. We built Skeptical to put reliability in your hands, whatever model you use.
Today, we’re out of stealth. skeptical.io
Paris Carbone retweeted
High availability is essential for AI and is typically achieved using distributed databases and KV stores. In my latest blog post, I show how fault-tolerance can instead be built directly into an AI chatbot using my Rust consensus library OmniPaxos.
omnipaxos.com/blog/building-…
Paris Carbone retweeted
Datacenter networking performance is becoming a critical bottleneck.
In our latest ADRS blog, we show how OpenEvolve automatically discovers improved congestion control by reducing queue length by 49% on the NSDI ’22 PowerTCP benchmark, starting from a baseline algorithm.
🎯 AI optimizes congestion control for datacenter networking
[ADRS Blog #13] We apply ADRS frameworks to improve congestion control algorithms. Using OpenEvolve, we discover an algorithm that reduces queue length by 49% on the NSDI ’22 PowerTCP 10:1 incast benchmark! 📉⚡️
✍️ Read the Blog: adrs-ucb.notion.site/cca
📖 ADRS Blog Series: ucbskyadrs.github.io/
📄 ADRS Paper: arxiv.org/abs/2510.06189
👩💻 Code: github.com/UCB-ADRS/ADRS
Paris Carbone retweeted
Systems research is changing: AI is enabling white-box algorithm discovery, not just black-box tuning.
Our new ADRS paper evaluates three frameworks and shows consistent gains over human designs—up to 13× speedup in load balancing for MoE experts and 35% lower cloud costs across different cloud regions.
🚀 Excited to share our updated paper on AI-Driven Research for Systems (ADRS)!
We evaluate 3 open-source frameworks across 10 real-world systems performance problems, showing that AI can generate solutions that outperform human experts, including a 13x speedup in load balancing and 35% cost savings in cloud scheduling.
🧵👇 Check it out!
Paris Carbone retweeted
Vol:19 No:1 → ConANN: Conformal Approximate Nearest Neighbor Search vldb.org/pvldb/vol19/p29-hor…
Paris Carbone retweeted
Vol:18 No:12 → Disaggregated State Management in Apache Flink 2.0 vldb.org/pvldb/vol18/p4846-m…
Paris Carbone retweeted
🏆 The ACM SIGMOD Systems Award 2025 goes to
Spanner: Google’s Globally-Distributed Database
"for reimagining relational data management to enable serializability with external consistency at global scale"
2025.sigmod.org/sigmod_award…
Paris Carbone retweeted
📢📢 Have a PhD position on the IO-path of deep learning, cofunded by Innovation Fund Denmark. Selected candidate will be affiliated with Samsung Research Denmark and @dasyaITU @ITUkbh. Supervised by @bonivivek, Javier González, & me. See details at linkedin.com/jobs/view/41416…
Paris Carbone retweeted
Hey Folks! 🌟
THE NEW ALBUM IS OUT today and it’s finally yours. It has been an epic 5 year journey until today and now it feels as if a whole world has changed during this period.
Produced by Tigran Hamasyan
@naiverecords
#tigranhamasyan #thebirdofathousandvoices
Paris Carbone retweeted
I am looking for full-time PhD students on topics centered around knowledge graphs spanning from graph data management to neurosymbolic AI and GraphRAG.
Application deadline: September 12, 2024
Link: jobs.tuwien.ac.at/Job/236402
Paris Carbone retweeted
This is a really cool exmaple of building a special index, for querying Parquet files using @ApacheDataFusio.
The research project makes SQL queries ~1000x faster with relatively straightforward optimizer pass in DataFusion and a novel index: uwheel.rs/post/datafusion_uw…
Paris Carbone retweeted
Excited to introduce datafusion-uwheel, a Rust crate that provides native DataFusion support for speeding up time-based analytics through µWheel.
Learn more: uwheel.rs/post/datafusion_uw…
Paris Carbone retweeted
"A survey on the evolution of stream processing systems"
Banger of a paper by @MarioFragkoulis, discussing ordering semantics, fault tolerance, scalability, state management and more, of different stream processors.
👉 link.springer.com/article/10…
Paris Carbone retweeted
µWheel 0.2.0 is now available! I cover both the new release and the recent DEBS conference experience in the following post - uwheel.rs/post/best-paper-aw…
Also, the paper is now available through the website uwheel.rs
Paris Carbone retweeted
I just realized that reviewer#2 used ChatGPT and didn't even bother to hide it. I appreciate the friendly and constructive feedback though.
Here's how I spot ChatGPT:
- all the points start with a statement written in title case (aka capitalised the beginning of big words) 1/2
Paris Carbone retweeted
In this post, I show the potential of integrating µWheel as an index within DataFusion (@ApacheDataFusio) to speed up temporal aggregation queries up to 60-60000x.
maxmeldrum.com/docs/posts/20…
Paris Carbone retweeted
Happy to share that our paper on µWheel, an Embeddable Aggregate Management System for Streams and Queries was accepted to DEBS24.
µWheel unifies the aggregate management for streaming window aggregation and adhoc queries over historical data.
Learn more github.com/uwheel/uwheel
Paris Carbone retweeted
We are happy to announce the release of our latest book titled “Programming Heterogeneous Hardware via Managed Runtime Systems” published by @SpringerNature together Juan Fumero (@snatverk) and Thanos Stratikopoulos (@thanos_str )!
link.springer.com/book/10.10…
Paris Carbone retweeted
If you’re in Athens for @EuroSys_conf checkout PAPOC, my favourite workshop for all things consistency. @ChrisJe34211511 will be talking about a hot new take on 2PC and David Chu will taking about programmatically scaling BFT protocols. Definitely worth a watch!
Excited for the awesome PAPOC'24 program at @EuroSys_conf in Athens, with keynotes from @WyattLloyd and Mohsen Lesani. Come here all about the latest developments in the consistency world!
Paris Carbone retweeted
On Valentine’s day, all you need is pip 😍
Whether you found your match in life or not, Valentine matches your datasets.
Feb. 14th anniversary-release (v0.2.0) includes:
— faster batch-matching
— more string similarity functions
— redesigned metrics
++
github.com/delftdata/valenti…
