Mikhail Rozhkov

Data Science, Machine Learning, Robotics

[Jay Chooi @chooi_jeq]

GPT-6 Astra scored 95% on a robot control task, up from Fable 5.1's 40%, with 6.2x fewer output tokens at 2.3x lower cost. 🧵

[Mikhail Rozhkov @mnrozhkov]

We built an OpenClaw example on Serverless Endpoints (CPU gateway + LLM via API) — kudos to Gleb Berjoskin 👏

Not optimal, but interesting experiment:
CPU → gateway
GPU → model
Repo: https://t.co/E48mIvdfe9
Curious how others run agent workflows — feedback welcome.

[Andrej Karpathy @karpathy]

Software horror: litellm PyPI supply chain attack.

Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords.

LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm.

Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks.

Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages.

Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

[Aaron Levie @levie]

https://t.co/O7OiUYKjbh

[Mikhail Rozhkov @mnrozhkov]

Looking for input from AI/ML practitioners 👇

Working on a Serverless Cookbook — runnable workloads:
• fine-tuning (Axolotl)
• LLM serving (vLLM/Qwen)
• simulations (OpenMM)

What would make these actually useful?
What should we add next?
https://t.co/E48mIvdfe9

[Nebius Token Factory @nebiustf]

⚡️Today marks a big milestone for Nebius.

We’re launching Nebius Token Factory, the evolution of Nebius AI Studio, built to make open-source AI production-grade.

Token Factory transforms raw open models into governed, scalable systems with dedicated inference, sub-second latency, 99.9% uptime and zero-retention compliance.

It’s where inference, post-training and governance converge, turning raw compute into reliable intelligence.

Run AI inference at scale: https://t.co/rRIK3rpaLI

[Nebius @nebiusai]

Tired of flying blind when developing LLM apps? Frustrated by mysterious agent behaviors? Can’t figure out why your agent fails in production?

You’re not alone. Register for the webinar on our website to learn how Managed @MLflow is a game-changer: https://t.co/4NBMySDkcc 1/4

[Mikhail Rozhkov @mnrozhkov]

Hey! Who is working on Agentic AI right now? Looking for best practices! Message me, let's talk! ✍🏻

Note: I'm preparing a webinar about Agentic AI development with Managed MLflow. Get your spot on April 17 ✅
https://t.co/f0EUZD6aCy

[Mikhail Rozhkov @mnrozhkov]

I'm pleased to present the ML System Design Canvas, which is the outcome of our collaboration with Alexander Guschin! https://t.co/zbmLGrcHf2

[Mikhail Rozhkov @mnrozhkov]

@DVCorg Yes! Finally, it's alive! I am looking forward to hearing the community's feedback! 🤓

[🦉DVC @DVCorg]

🦉Ready to level up your AI development game? Learn how to use the power of @DVCorg and @raydistributed (from @anyscalecompute ) to streamline the complexity of your ambitious projects without sacrificing efficiency in this 2-part tutorial from @mnrozhkov.

In these blog posts, you will learn how DVC automates pipelines, manages versions, and tracks experiments, all while embracing GitOps for reproducibility. And how to combine that with Ray to scale distributed computing and training.

Part 1 covers:

🔹 Why DVC and Ray
🔹 Tutorial Scope
🔹 High-level solution design
🔹 Installation
🔹 Getting Started with Ray
🔹 Run DVC Pipeline on a Ray Cluster
🔹 Discussion of the Solution Design
🔹 Sum up - DVC + Ray Integration

Part 2 covers:

🔹 Design Scalable ML Experiments with DVC & Ray
🔹 Set up and Rund DVC in Distributed Ray Cluster on @awscloud (AWS)
❍ Prepare AWS and DVC Studio credentials
❍ Configure Ray Cluster in cluster.yaml
❍ Start a Ray Cluster on AWS
❍ Connect to the Head Node and Set up Credentials
❍ Run DVC Pipelines on the remote Ray Cluster
❍ Commit & Push Experiments
❍ Stop Cluster 🛑

There are lots of details and code in the blog posts for ML development at scale satisfaction!

Part 1 👉🏽 https://t.co/z0MTnGvWMj
Part 2 👉🏽 https://t.co/qaxxN9tWz6

[Mikhail Rozhkov @mnrozhkov]

Hey! I'm happy to share some valuable insights on DVC and Ray integration!

Tutorial: Scalable and Distributed ML Workflows with DVC and Ray
- Part 1: https://t.co/4iEv1oifbC
- Part 2: https://t.co/kNNhvakyUc

#ai #ml #mlops #dvc #ray #ray

[Mikhail Rozhkov @mnrozhkov]

It's never too late to start a new Medium :))) Just about the latest tutorial with DVC and Evidently https://t.co/Oi9PyY6HB0

[Mikhail Rozhkov @mnrozhkov]

Recently, our team published a new post about integrating Evidently and DVC. I'm excited to hear your thoughts on this!
https://t.co/CC0B7ZSoa0

#dvc #evidently #mlops

[Mikhail Rozhkov @mnrozhkov]

🎉 Congratulations to my team for launching their amazing DVC extension for VS Code on Product Hunt!
GitOps ML just got a whole lot friendlier with this gem! 🚀🤖🪙

Support us - add your thoughts on PH 🙏

Let's show some support – check it out here: https://t.co/mnPrWFmW3v

[Machine Learning REPA @mlrepa]

This week at ML REPA, we're organizing an online meetup-workshop titled "Talking with AI."

➡️ Learn how build and use your own AI Agents

Join us: Event Link https://t.co/UvxrxAhgIw 🤖📣

[🦉DVC @DVCorg]

@mnrozhkov discusses the landscape of features and use cases of DVC in ML engineering and MLOps

Learn more here - https://t.co/OSbANBESx7

[Mikhail Rozhkov @mnrozhkov]

Thank you, Elena, and the entire Evidently AI team! I'm thrilled about the opportunity to collaborate and explore new integrations and solutions together. Let's continue creating valuable tutorials that benefit the community! 🤝 https://t.co/5PP6pWhNph

[Evidently AI @EvidentlyAI]

How to set up an ML monitoring system for models deployed with @FastAPI?

⚡️ Create a FastAPI application to serve predictions
🔢 Log them to a PostgreSQL database
📊 Create an Evidently monitoring dashboard to track ML model performance

Code tutorial 👇
https://t.co/jErJIyufip

[Machine Learning REPA @mlrepa]

🎉 We are thrilled to announce the launch of our free community course, git-101: "Git for Machine Learning and AI Development" at ML REPA School! 🌟

🌟 Join us at ML REPA School and unlock the power of Git for your ML and AI projects today! https://t.co/REuJwFPsEA

#mlrepa #ml