Lead Software Engineer 👨💻 - Building Data and AI Apps - Simplifying Python, Data Science, ML and LLMs - itsafiz@gmail.com for collab - Opinions are my own
- Tweets 12,418
- Following 547
- Followers 42,990
- Likes 17,149
Chat models generate messages.
Decision models make decisions.
Workflow models execute entire workflows in a single API call.
Stop writing orchestration loops around your LLM.
TypeLLM is a workflow model. You declare the decisions, their types and what depends on what. It runs the whole graph in ONE API call.
Think SQL for AI workflows.
Support ticket example:
→ Classify category + urgency in parallel
→ Refund only runs if billing
→ Reply reads category + urgency
Four steps. One call. Typed results.
vs four serial calls:
→ 2.1x fewer input tokens
→ ~1.9x faster
4
3
9
3,233
Afiz ⚡️ retweeted
Don't waste your time, just use graphify skill to turn your codebase into an interactive map.
Now available to all @perplexity_ai
Computer users for FREE.
Computer
The Graphify skill by @graphify is now in Computer's skills marketplace, available to all Computer users.
Turn a codebase into an interactive map of its files, functions, and dependencies.
Tasks that use /graphify are free through October 31, up to 5,000 credits per user.
2
10
1,283
Afiz ⚡️ retweeted
Now you can run Perplexity Computer fully local on your Mac!
No wifi needed.
works with Mac apps and 100% local.
Perplexity
You can now run Perplexity Computer fully local on your Mac with no cloud based inference, without WiFi too. Enjoy!
4
2
9
1,996
Read more:
cloud.google.com/blog/produc…
2
107
Afiz ⚡️ retweeted
You're deep in a content draft when you remember a client needs a follow-up.
You won't stop to open a reminders app. You'll tell yourself you'll remember. Then you don't.
That's where SpeakON makes sense to me. Say it out loud, and the agent helps set the reminder.
Voice Keyboard and Notes capture the thought. Agent helps you act on it.
The useful part isn't voice typing. It's going from saying something to getting help with the next step, without breaking focus.
For anyone juggling clients and content, that gap is where things slip.
we crushed a keyboard.
its retirement party got a little out of hand.
for years, it was there for every email, every plan, every “just one more thing.” a loyal colleague. a very loud colleague.
but your ideas move faster than your fingers.
and somewhere between knowing what you want to do and typing it all out, a quick task becomes another thing taking up your afternoon.
meet SpeakON AI Agent.
start with what you need, in your own words. less time behind a blinking cursor. more room to move things forward.
the keyboard had a good run.
now, let’s give your voice a turn.
THE LAST KEYSTROKE is here. watch the film, then tell us: what’s the first task you’d hand over?
speakon.app/x
Disclaimer: The calling feature is still in beta and will be available soon. All other Agent features are available now, with ongoing improvements.
2
6
1,123
The team behind Google's Gemini and Apple's MM-1 just released their first model.
Meet Glimpse from @ElorianAI 👁️
Text models went superhuman at code. Images and video are still hard.
Elorian is a whole lab built for Visual Thinking, and Glimpse is the first result.
→ Reasons over charts
→ Counts under trick conditions
→ Understands why a video clip is funny
→ Labels every subtask in robotics videos
Free chat access. API is live.
Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that.
Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.
3
3
7
1,066
What to learn How do always-on, proactive agents work?
Here is the great lecture at Stanford CS146S by @leerob
Lee Robinson
1
3
3
817
You might want to check out this post on workflow models.
Chat models generate messages.
Decision models make decisions.
Workflow models execute entire workflows in a single API call.
Stop writing orchestration loops around your LLM.
TypeLLM is a workflow model. You declare the decisions, their types and what depends on what. It runs the whole graph in ONE API call.
Think SQL for AI workflows.
Support ticket example:
→ Classify category + urgency in parallel
→ Refund only runs if billing
→ Reply reads category + urgency
Four steps. One call. Typed results.
vs four serial calls:
→ 2.1x fewer input tokens
→ ~1.9x faster
2
123
Chat models generate messages.
Decision models make decisions.
Workflow models execute entire workflows in a single API call.
Stop writing orchestration loops around your LLM.
TypeLLM is a workflow model. You declare the decisions, their types and what depends on what. It runs the whole graph in ONE API call.
Think SQL for AI workflows.
Support ticket example:
→ Classify category + urgency in parallel
→ Refund only runs if billing
→ Reply reads category + urgency
Four steps. One call. Typed results.
vs four serial calls:
→ 2.1x fewer input tokens
→ ~1.9x faster
4
3
9
3,233
Quick start TypeLLM for FREE: typellm.ai/docs/quick-start
1
3
97
I hope you've found this thread helpful.
Follow me for more such content.
Like/Repost the quote below if you can:
Chat models generate messages.
Decision models make decisions.
Workflow models execute entire workflows in a single API call.
Stop writing orchestration loops around your LLM.
TypeLLM is a workflow model. You declare the decisions, their types and what depends on what. It runs the whole graph in ONE API call.
Think SQL for AI workflows.
Support ticket example:
→ Classify category + urgency in parallel
→ Refund only runs if billing
→ Reply reads category + urgency
Four steps. One call. Typed results.
vs four serial calls:
→ 2.1x fewer input tokens
→ ~1.9x faster
3
90
Afiz ⚡️ retweeted
Embedding Gemma 2 by @Google is trending on @huggingface!
Well deserved place for Embedding Gemma 2.
3
15
2,498
Google DeepMind is giving 10,000 scientists free highest-tier access to Antigravity for 6 months.
This is the agent-first coding platform now running Gemini 4 Argon. Solid chance to push real research workflows without the usual usage limits.
Apply if it fits your work.
We are offering 10,000 scientists the highest usage tier in Google Antigravity free of charge for 6 months. Please apply! forms.gle/WvmBbPvRbpKG1BPn9
3
4
8
2,453
Google DeepMind is giving 10,000 scientists free highest-tier access to Antigravity for 6 months.
This is the agent-first coding platform now running Gemini 4 Argon. Solid chance to push real research workflows without the usual usage limits.
Apply if it fits your work.
We are offering 10,000 scientists the highest usage tier in Google Antigravity free of charge for 6 months. Please apply! forms.gle/WvmBbPvRbpKG1BPn9
3
4
8
2,453
Apply Here: docs.google.com/forms/d/e/1F…
1
205
Gemini 4 Argon is here.
Google is using Argon on its hardest engineering problems.
- Chrome: untangled a ~6M line legacy folder across hundreds of files.
- Gemma 4 E2B: 2.85x faster inference on an A100 via vLLM.
More details 👇
Sundar Pichai
1
2
6
3,087
Gemini 4 Argon is here.
Google is using Argon on its hardest engineering problems.
- Chrome: untangled a ~6M line legacy folder across hundreds of files.
- Gemma 4 E2B: 2.85x faster inference on an A100 via vLLM.
More details 👇
Sundar Pichai
1
2
6
3,087