Pinned Tweet
Want to talk about data?
Data science, career questions, practical machine learning, and more?
Join @DataTalksClub! The link is in the profile description.
Learn more about our activities in the thread! 🧵
Data science, career questions, practical machine learning, and more?
Join @DataTalksClub! The link is in the profile description.
Learn more about our activities in the thread! 🧵
1
8
42
17,199
DataTalksClub retweeted
What breaks first when you put an image, video, or audio model into production?
- Cost
- Latency
- Failures: out-of-memory, timeouts, killed jobs
- Quality: is the output good enough?
Reply and tell us.
- Cost
- Latency
- Failures: out-of-memory, timeouts, killed jobs
- Quality: is the output good enough?
Reply and tell us.
1
2
2
376
The materials for Production AI Agents for Generative Media are free for everyone. Homework, the final project, and peer review are not.
Those are for selected participants only. Here is what we look at in your application:
Those are for selected participants only. Here is what we look at in your application:
1
1
1
350
Detailed answers help. Applications filled in superficially probably won't get in.
The course is for ML, MLOps, and data engineers, backend engineers, and founders working with media generation.
The course is for ML, MLOps, and data engineers, backend engineers, and founders working with media generation.
1
0
0
32
DataTalksClub retweeted
Build or buy? Autotrader ran both on live traffic.
Build: Spark Streaming, tens of seconds, thousands of lines of code
Buy: Snowplow Identities, milliseconds
1/3
Build: Spark Streaming, tens of seconds, thousands of lines of code
Buy: Snowplow Identities, milliseconds
1/3
1
2
1
356
DataTalksClub retweeted
Connecting an AI agent straight to your data warehouse looks like the quickest way to build an analytics agent. It fails.
On Monday, Alex Litvinov from @ManychatHQ will explain why, and what sits between the agent and the data instead.
On Monday, Alex Litvinov from @ManychatHQ will explain why, and what sits between the agent and the data instead.
1
1
2
224
DataTalksClub retweeted
An image generation script that works locally isn't a GPU Job yet.
To run it on serverless GPUs, the script needs to run in a container, which raises questions that a local run never does.
1/4
To run it on serverless GPUs, the script needs to run in a container, which raises questions that a local run never does.
1/4
1
2
1
375
What breaks first when you put an image, video, or audio model into production?
- Cost
- Latency
- Failures: out-of-memory, timeouts, killed jobs
- Quality: is the output good enough?
Reply and tell us.
- Cost
- Latency
- Failures: out-of-memory, timeouts, killed jobs
- Quality: is the output good enough?
Reply and tell us.
1
2
2
376
Our new free course, Production AI Agents for Generative Media, covers AI agents that run these models in production and covers all four.
Materials are free for all. But the live cohort is for selected participants only.
Learn more here: github.com/DataTalksClub/…
Materials are free for all. But the live cohort is for selected participants only.
Learn more here: github.com/DataTalksClub/…
0
0
1
38
DataTalksClub retweeted
Alexey Grigorev built a Pydantic AI agent that turns a child profile into a personalized Christmas illustration. Mikhail Rozhkov showed the GPU endpoints behind it.
Watch our workshop recording:
youtube.com/watch?v=XzS6Mr…
1/5
Watch our workshop recording:
youtube.com/watch?v=XzS6Mr…
1/5
1
3
3
577
Build or buy? Autotrader ran both on live traffic.
Build: Spark Streaming, tens of seconds, thousands of lines of code
Buy: Snowplow Identities, milliseconds
1/3
Build: Spark Streaming, tens of seconds, thousands of lines of code
Buy: Snowplow Identities, milliseconds
1/3
1
2
1
356
Stitching mattered most for shoppers close to a purchase.
Watch the session:
insights.snowplow.io/l/1092822/2026…
2/3
Watch the session:
insights.snowplow.io/l/1092822/2026…
2/3
1
0
0
33
This post is made in collaboration with Snowplow. Thank you for supporting our community!
3/3
3/3
0
0
0
27
Connecting an AI agent straight to your data warehouse looks like the quickest way to build an analytics agent. It fails.
On Monday, Alex Litvinov from @ManychatHQ will explain why, and what sits between the agent and the data instead.
On Monday, Alex Litvinov from @ManychatHQ will explain why, and what sits between the agent and the data instead.
1
1
2
224
He'll cover context layers, how to keep that context fresh as the business evolves, and how to make answers verifiable, with a human in the loop.
1
0
0
43
DataTalksClub retweeted
Not sure whether Production AI Agents for Generative Media is the right course for you? Ask before you apply.
On November 10, Alexey Grigorev will answer questions about our new free course live on YouTube.
1/4
On November 10, Alexey Grigorev will answer questions about our new free course live on YouTube.
1/4
1
2
2
383
An image generation script that works locally isn't a GPU Job yet.
To run it on serverless GPUs, the script needs to run in a container, which raises questions that a local run never does.
1/4
To run it on serverless GPUs, the script needs to run in a container, which raises questions that a local run never does.
1/4
1
2
1
375
Module 1 of Production AI Agents for Generative Media, our new free course with Nebius, covers this part: containers and model cache for media models on serverless GPUs, plus seeds and fitting VRAM.
Learn more here: github.com/DataTalksClub/…
3/4
Learn more here: github.com/DataTalksClub/…
3/4
1
0
0
49
Alexey Grigorev will introduce Module 1 at the launch stream, the day it starts.
📅 Monday, November 16, 17:00-18:00 CET, live on YouTube
Register: luma.com/xey7h50c
4/4
📅 Monday, November 16, 17:00-18:00 CET, live on YouTube
Register: luma.com/xey7h50c
4/4
0
0
0
27