Milind Srivastava

PhD @CarnegieMellon | @iitmadras CS

[Laude Institute @LaudeInstitute]

Introducing Slingshots // FOUR: 14 projects across eight top institutions. Self-improving agent systems and harnesses, maximally-open models and training data, applications for research and decision-making – plus kernels and more. Say hello to the batch 🧵

[vyas sekar @vyas_sekar]

While AI-evolved optimized systems may well work in the long term, current versions do not appear to be ready for primetime.

We discuss this in a new blog: https://t.co/EvuxeFKgkn

What you can do: do not blindly trust AI-optimized systems . Explicitly stress test them!

[vyas sekar @vyas_sekar]

Recent work by @CyLab students shows that open-weight models have become very capable autonomous multi-host attackers: https://t.co/oMWd0gVCG7.

This is the first of our blog series from Cyber Autonomy Initiative https://t.co/oZ0MYd5IF3 @CyLab

[Vinod Ganesan @vinodgansan]

Alma - Reliable, fast, and cost efficient computer use for all. Coming soon!

[vyas sekar @vyas_sekar]

Congrats to @CMU_ECE PhD student Steven for winning the ACM SIGCOMM SRC. Here is a nice description of the award-winning work https://t.co/MyZ6L2GaxD

[Omer Akgul @_oakgul]

Heading to USENIX Security tonight. Would love to catch up with folks that are around.

Also, we're hiring @incalmo_ai, come talk to me!

[Milind Srivastava @milindsriv]

@spice_ai @ApacheDataFusio @sap1ens Thanks for having me @spice_ai !

[Spice AI @spice_ai]

Great @ApacheDataFusio Community Showcase yesterday with presentations from @milindsriv (ASAPQuery/Carnegie Melon University) and @sap1ens (Streamling).

Watch the recording here: https://t.co/UE9IvSlqSL

[Spice AI @spice_ai]

Join the third edition of the @ApacheDataFusio Community Showcase this Thursday, August 6th at 3 pm PST on Zoom!

The Showcase will feature two DataFusion builders:

1. "ASAPQuery: A Drop-in Sketch-based Query Accelerator for Metrics Observability", presented by Milind Srivastava

​ASAPQuery is a drop-in query accelerator that fits in your existing metrics observability stack like Prometheus-Grafana or Clickhouse-Grafana. ASAPQuery uses sketch-based approximation to reduce query latency by 2 orders of magnitude while simultaneously reducing query cost. Internally, ASAPQuery converts existing PromQL/SQL queries to Datafusion-based plans (with custom nodes) and executes them using sketches.

2. "​Streamling: Building a Lightweight Data Streaming Runtime on Apache DataFusion", presented by Yaroslav Tkachenko

​Many analytical databases and query engines use DataFusion as a foundation. But can you extend it to build a data streaming framework? In this talk, Yaroslav will cover the lessons learned from extending DataFusion to support streaming. We'll look into how well most of the existing components can be leveraged with little or no modifications. We'll also see how modern features, like implementing TypeScript support via WebAssembly, can be integrated with the Arrow format.

cc @andrewlamb1111

Register here ➡️ https://t.co/obf3I6Ukd8

[Brian Singer @BrianSinger98]

The future is that all malware will be completely AI generated

[vyas sekar @vyas_sekar]

The incident from @huggingface and @OpenAI (https://t.co/kLUjo3AOXQ https://t.co/j9AqA2fs5U) reconfirms the argument we have been making at @CyLab for a Scientific Foundation for Cyber Autonomy (https://t.co/yUEXGa4A75)

Reach out to us to collaborate!

[Milind Srivastava @milindsriv]

@moarbugs @AntithesisHQ Congrats Rohan!

[Laude Institute @LaudeInstitute]

Introducing Slingshots // THREE: 13 projects from the top researchers and top institutions in US and Canada. Alt architectures, world models, continual learning, hard sciences, HCI, and more. Let's meet the batch 🧵

[Milind Srivastava @milindsriv]

I recently presented ProjectASAP's ASAPQuery at GrafanaCon. My first time there, both as an attendee and speaker. Probably the largest audience I've presented to so far. I met a ton of new people and had insightful conversations about ASAPQuery, Grafana and observability.

[Milind Srivastava @milindsriv]

I spoke at at Observability Day Europe about how ASAPQuery (https://t.co/EwCSnu36O1) can make metrics observability queries 100x faster while simultaneously reducing query CPU and memory cost.

Check out the talk video -- https://t.co/eBjP1cUZ4P

[Milind Srivastava @milindsriv]

@CarnegieMellon @UofMaryland @LaudeInstitute @Alan_Lau @vyas_sekar @zeying_zhu

[Milind Srivastava @milindsriv]

Grafana dashboards timing out on large Prometheus deployments? We built ASAPQuery, a drop-in accelerator for Prometheus-Grafana that speeds up queries by 100x!

Run the 5-min Docker quickstart demo to compare Prometheus vs ASAPQuery latency side-by-side: https://t.co/EwCSnu36O1

[Milind Srivastava @milindsriv]

ASAPQuery is born out of research at @CarnegieMellon and @UofMaryland University of Maryland.
Thanks to @LaudeInstitute for supporting this work!

[Milind Srivastava @milindsriv]

Thanks to Laude for supporting our work. Stay tuned for something exciting in data analytics and observability!

[Laude Institute @LaudeInstitute]

ASAPQuery/ @milindsriv @Alan_Lau @zeying_zhu @vyas_sekar / (@CarnegieMellon) This infrastructure replaces expensive and slow “scan everything” data analysis with lightweight summaries, so teams (and AI agents) can get useful answers from massive datasets much faster and at lower cost.