I create content about deep learning | The DevRel and Deep Learning Dude

I don’t wanna hop on call to learn about or see a demo of what you’re building. Too busy hacking on VLMs, curating multimodal datasets, and contributing to the open source community. 

Harpreet Sahota 🥑

It was great to be back in Boston last week and host another awesome meetup!

Thank you for everyone who showed up, our awesome lineup of speakers for taking time out of their schedules for delivering such informative talks: Anku Rani Hanhui Wang Sage Simhon Sharut G. 

And for Voxel51 for sponsoring!

If you missed the event, checkout the comments for where you can find the recordings of the talks

🤖 Hacker-in-Residence @ Voxel51| 👨🏽‍💻 AI/ML Engineer | 👷🏽‍♀️ Technical Developer Advocate | Learn. Do. Write. Teach. Repeat.

No alternative text description for this image

i built a ComfyUI-inspired visual, node-based workflow editor for FiftyOne

now you can build dataset curation, analysis, and model evaluation pipelines by connecting nodes on a canvas, without writing code

give it a try and let me know what you think, completely open source and all you have to is `pip install fiftyone` and follow the instructions on the readme

everyone is hype about gtc, but the 3d vision conference is this week and i'm much hype for that

i made a repo and a dataset to explore the papers

checkout the repo here:
 https://lnkd.in/gRFz6Kb7

here's a dataset that you can use to explore the papers:
 https://lnkd.in/g2RA87fD

it's been a long, long while since i've done a live stream on linkedin, but it's happening in t-mius 8 minutes!

if you're interested in learning about how to work with video datasets using open source tools, open data, and open models...this is the session for you!

Join us on Feb 26 for Exploring Video Datasets with FiftyOne and Vision-Language Models. Register for the Zoom: https://lnkd.in/gCk7gGPi

Video is the hardest modality to work with — more data, more complexity, and annotation workflows that don’t scale. This hands-on session by Harpreet Sahota 🥑 shows how to understand what’s inside massive video datasets without manually watching thousands of clips.

You’ll learn how to:
🎬 Navigate and explore video datasets in the FiftyOne App and understand dataset structure
🔎 Compute embeddings with Qwen3-VL to enable semantic search, zero-shot classification, and clustering
📝 Generate video descriptions and localize events using vision-language models like Qwen3-VL and Molmo2
🧭 Visualize patterns and trends through embedding projections in FiftyOne
📏 Evaluate model outputs against Action100M’s hierarchical annotations

Perfect for ML engineers, data scientists, and computer vision teams working with large-scale video data who want practical workflows for exploration, curation, and debugging.

#computervision #machinelearning #ai #datascience #videoanalytics #visionlanguagemodels #mlops #opensource

Derek Van de Streek, Swati Bannore Jain and 11 other attendees

Feb 26 - Exploring Video Datasets with FiftyOne and Vision-Language Models

excited to be back in stuttart! if you're in the area and want to speak at the meetup on april 21st, send me a dm

Chief Community Officer @ Voxel51 | Startup Advisor

Are you based in the Stuttgart area? Consider speaking at the upcoming April 21 - Stuttgart AI, ML and Computer Vision Meetup: https://voxel51.com/speak

#computervision #ai #artificialintelligence #machinevision
#machinelearning #datascience #opensource #stuttgart

i've learned a lot from Eric Ma over the years, from the first time I saw one of his talks on principled data science workflows to talks he's given for my communities over the years, and his latest work on getting the most from coding agents.

check out his next talk, and support the SciPy community as well!

Together with my teammates, we solve biological problems with network science, deep learning and Bayesian methods.

As a data scientist, how much of your week is spent on work a coding agent could do?

I once planned multiple days of work for a machine learning experiment last year. This was an experiment that I had put off for an entire quarter because, as a team lead, I didn't have the time to block out the 3-4 afternoons that I thought I'd need. 

Instead, I described the problem to Claude Opus 4.5 via OpenCode, and it wrote the code, ran the training, found issues, switched strategies, and finished the whole thing in 20 minutes, complete with a written report drafted *with* figures referenced.

The data scientists who figure this out now will have an edge.

I'm teaching a workshop on how to do this flavour of agentic data science to raise funds for SciPy 2026's Financial Aid Program. You'll learn:

- Prescriptive prompting for experiments
- Project structure for agents
- Logging and verification strategies to enable automated control
- Report writing with AI
- Staying in control during EDA
- The minimalist version first

These are the same techniques I use in my personal projects and my teammates to compress data science timelines from days to minutes. I think you'll have a chance to learn something good while doing something good!

Details:

- Date: March 27 & April 3, 2026
- Time: 2:00-4:00 PM EST
- Venue: Google Meet (calendar invite will be provided)
- 40 spots available
- $1,000 minimum donation, paid directly to NumFOCUS (tax-deductible), student pricing: $200!

Video recordings available if you can't attend live.

Sign up here: https://lnkd.in/eFjc36Tb. I'm looking forward to seeing what you'll build with these techniques! 

P.S. If you can't make it but share it with others who have the means and would benefit from the workshop, I would deeply appreciate it!

Open article: Agentic Data Science Workshop | SciPy 2026 Fundraiser by agent-assisted-data-science.vercel.app

Agentic Data Science Workshop | SciPy 2026 Fundraiser

We just released a new AWS workshop that I’m really excited to finally share 👇
https://lnkd.in/eiNprTd7

This came from something I keep seeing across autonomy, robotics, and computer vision teams lately.

Most teams are not blocked by models anymore.
They are blocked by data and iteration speed.

So we wanted to show what a practical workflow actually looks like:
🚗 generate synthetic driving data
 🧠 auto label using foundation models
 🔧 fine tune a small vision language model
 ☁️ deploy and iterate using AWS SageMaker
 👀 inspect and improve everything inside FiftyOne

The goal was simple.

Build something real that teams could follow end to end.

Huge credit to Ryan Patrick Sweeney, Mark Vinciguerra, and Derick Thompson for driving this forward, along with great collaboration between Amazon Web Services (AWS), NVIDIA Omniverse, and Voxel51 🙌

Also excited that Voxel51 is now live on AWS Marketplace, which makes getting started a lot easier for teams working on computer vision and robotics workflows!

Personally, I think this is how a lot of Physical AI systems will get built going forward.

Faster loops between data, models, and deployment instead of long waiting cycles.

If you’re working on robotics, AV, or multimodal systems, I’d genuinely love to hear what you’re building.

More coming soon 🚀

#PhysicalAI #ComputerVision #Robotics #MultimodalAI #SyntheticData #AWS #MachineLearning #AutonomousSystems

Open article: GitHub - aws-samples/sample-finetuning-small-vlms-synthetic-data by github.com

Contribute to aws-samples/sample-finetuning-small-vlms-synthetic-data development by creating an account on GitHub.

GitHub - aws-samples/sample-finetuning-small-vlms-synthetic-data

I’m so happy to have my boi Wiz present at one of our meets!

AI @ NVIDIA - 🏗️ Build | 🚢 Ship | 🚀 Share

View Chris "The Wiz 🪄" Alexiuk’s profile

skills.md - you've seen it, you've probably used it (I assume if you're following this LinkedIn account, you're someone who uses tools like OpenCode/etc.

I'm excited to be yappin' about skills at an upcoming Voxel51 / Hugging Face virtual event with a line up that boggles the mind: 

Ben Burtenshaw, Daniel Ávila Arias, and AdonAI Vera.

My presentation will focus on something near and dear to my heart: Skills as documentation. 

The days of documentation aren't over - humans, and agents, still need to understand how things work - but they're changing quickly. Not everyone wants to read a thorough getting started guide that may be built for a specific system (and if it's not, it'll be ~2K words long) Sometimes, you just want to clone a repository - and get cooking. 

Skills help bridge that gap - context for LLMs, but tailored for a human experience. 

Can't wait for the event, come on out and ask any skill questions you have!

looking forward to hosting this workshop tomorrow, join me!

Join Harpreet Sahota 🥑 on Feb 26 for a virtual workshop to learn how to use Facebook's Action100M dataset and FiftyOne to build an end-to-end workflow - https://lnkd.in/gFvwbQni

What you'll learn:

* Navigate and explore video data in the FiftyOne App, filter samples, and understand dataset structure

* Compute embeddings with Qwen3-VL to enable semantic search, zero-shot classification, and clustering

* Generate descriptions and localize events using vision-language models like Qwen3-VL and Molmo2

* Visualize patterns in your data through embedding projections and the FiftyOne App

* Evaluate model outputs against Action100M's hierarchical annotations to validate what the models actually capture

By the end of the session, you'll have a reusable toolkit for understanding any video dataset at scale, whether you're curating training data, debugging model performance, or exploring a new domain.

***********

Want to build better computer vision models? FiftyOne is an open source toolkit from Voxel51 (our Meetup sponsor) that helps you curate datasets, evaluate model performance, visualize embeddings, catch annotation errors, and eliminate duplicate images—all in one place.

“pip install fiftyone” is all it takes to get started - https://docs.voxel51.com/

#computervision #ai #artificialintelligence #machinevision #machinelearning #datascience #opensource

No BS AI/ML Content | ML Engineer with a Plot Twist 🥷30M+ Views 📝

Done-for you Audience Building & Content Writing for AI/ML Founders that frees up 100% of your time, so you can focus entirely on building your company  
  
Is your work too important to ignore? 💌 https://tally.so/r/wk69VJ  (or shoot me a DM) 👉
  
It's 2025 and we're at the peak of the attention economy  
  
Where Authenticity is king, visibility is currency and people wanna buy from people

Not faceless dehumanized brands  
  
So, they told you LinkedIn was "free marketing" you slapped some posts and called it strategy  
  
You've poured hours crafting the 'perfect' post, only to watch it sink without a trace..  
  
Then it dawned on you, your technical brilliance isn't enough to draw a crowd  
  
Hell, maybe you've even tried mimicking influencers, scrambling for any traction..  
  
Your post got 500 f*cking views... from your friends...  
  
Being ignored is soul-crushing  
  
It's not your fault tough  
  
When you started your company, you didn't sign-up for a degree in an Audience Building, Copywriting and LinkedIn strategy 

And here you are, sacrificing yet another late night for your vision  

"Am I just sabotaging my efforts?"  
  
"Is my life-work meaningless?"  
  
Getting traction feels impossible  
  
You hoped hiring some marketer would save your weekends but..  
  
Tech is a code they can't crack  
  
They can only churn out the lame, soul-sucking crap we all ignore on sight  
  
With armies of founders flooding feeds, your voice dissolves into meaningless white noise  
  
But in the process, you've wasted more of your precious time explaining 101 tech concept and rewriting low-quality drafts 

Done being ignored?  
  
Let's make the world listen 
  
I can help you with:  

✒️ No BS AI/ML Content  
⭐ LinkedIn Profile Revamp
🎯 Website Copy Optimization  

I've done it before:

↳ For founders backed by YC, a16z, Sequoia, and Techstars
↳ For bootstrappers in LLMs, ML, Vector DBs, Data Analytics, and Engineering
↳ For AI and Machine Learning leaders at Google, Microsoft, Airbnb, and YouTube

If they could rise above the noise, so can you

Yeah, even if you're an introvert with a "boring" niche

Still with me?  
  
Amazing—You're amazing!  
  
Cheers,  
Paolo

My booking link: 💌 https://tally.so/r/wLW0oz (or shoot me a DM) 👉

Paolo Perrone

LangChain just open-sourced their answer to Claude Code.

9.9K GitHub stars. 5 hours old.

Deep Agents: the batteries-included agent harness. MIT licensed. FREE.

Here's why this is a big deal:

Claude Code is amazing but closed-source and locked to Anthropic.

Deep Agents gives you the same architecture. Any model. Any provider.

What's included out of the box:

1️⃣ Planning
write_todos for task breakdown and progress tracking. The agent plans before it codes.

2️⃣ Filesystem access
read_file, write_file, edit_file, ls, glob, grep. Full codebase navigation.

3️⃣ Shell execution
Run commands with sandboxing. Build, test, deploy from the agent.

4️⃣ Sub-agents
Spawn isolated workers with their own context windows. Parallel work without context pollution.

5️⃣ Context management
Auto-summarization when conversations get long. Large outputs saved to files. The problem Claude Code solved, now OPEN SOURCE.

Two lines to start:

pip install deepagents

agent = create_deep_agent()

Works with Claude, OpenAI, Google, or any LangChain-compatible model. Swap providers without rebuilding.

Built on LangGraph. Streaming, persistence, checkpointing: all production-ready.

The CLI adds web search, remote sandboxes, persistent memory, custom skills, and human-in-the-loop approval.

From their README: "This project was primarily inspired by Claude Code."

They're not hiding it. They're DEMOCRATIZING it.

Are you still locked into a single AI coding provider? 👇

♻️ Repost for someone who wants Claude Code capabilities without the lock-in

No BS AI/ML Content | ML Engineer with a Plot Twist 🥷100M+ Views 📝

My Meta Staff Engineer friends shared their Claude Code setup with me.

Changed how I build software FOREVER.

10 tips. Here's why they matter:

1️⃣ Always start in Plan Mode
Never execute first. Argue with Claude. Challenge its assumptions. Spend 80% of your time here. The code generation is the easy part.

2️⃣ Run /init on every project
Claude scans your codebase and creates a CLAUDE.md rules file. This controls everything. Skip this and Claude is guessing blind.

3️⃣ CLAUDE.md under 300 lines
Every line bloats your context window. More bloat = worse output. Only add what's unique to your project. Think lint rules, not docs.

4️⃣ Build a validation loop
Build, test, compile commands inside CLAUDE.md. Claude self-corrects when it can validate its own output. Without this, it hallucinates fixes.

5️⃣ Interrupt early, interrupt often
Claude going off track? Escape. Don't wait. Don't hope. Fresh context beats bloated context every single time.

6️⃣ /clear between features
Old context contaminates new tasks. One feature per session. Start clean or pay the token tax.

7️⃣ Multiple instances simultaneously
This is the REAL unlock. Multiple terminals. Multiple Claude sessions. One builds, one plans, one debugs. Jump between them like Starcraft.

8️⃣ Never manually edit rules
"Update the rules so we never do this again." One sentence. Claude maintains its own CLAUDE.md. Stop hand-editing config files.

9️⃣ Audit with /context regularly
MCPs destroy your token budget. /context shows the biggest offenders. Disable what you don't need. Most engineers never check this.

🔟 Sub-agents for atomic tasks ONLY
Don't create CEO agents and product agents. Keep shared context in ONE session. Sub-agents lose the reasoning chain. Bring work to the context, not context to the work.

The meta-lesson:

AI coding is 80% context engineering, 20% code generation.

Most engineers optimize the 20%.

These 10 tips optimize the 80%.

That's why they ship faster than entire teams.

Are you still treating Claude Code like a chatbot? 👇

♻️ Repost for someone using Claude Code who hasn't optimized their workflow

2 founders built Claude Code but for marketing, and it's already compressing 10-month execution cycles into 10 days

After multi-billion dollar exits to Salesforce and Oracle, Tom Chavez and Vivek Vaidya noticed something:

AI eliminated manual work for engineers, but marketers were still dealing with analysis and dashboards.

I was watching their conversation with Navin Chaddha from Mayfield and it struck me what has driven their success as serial entrepreneurs:

•⁠ ⁠“A bad decision made quickly is better than a good decision made slowly.”
•⁠ ⁠“Everything we used to do in 10 months, we now try to do in 10 days.”

Hear them break it all down:

Everyone's building AI agent demos.

Palantir shipped an end-to-end agentic architecture.

12 layers. The ENTIRE production stack. Here's what's inside:

1️⃣ Secure LLM integration
GPT, Gemini, Claude, Llama, Meta: commercial and open-source. Zero data retention by providers. Swap models without rebuilding.

2️⃣ Context Engineering
Real-time, pro-code and no-code. Contextual data, contextual logic, and systems of action feeding the Ontology.

3️⃣ The Ontology
Not a data lake. A digital twin with semantic, kinetic, and dynamic layers. Human+AI decision model built in. This is their moat.

4️⃣ Agent Lifecycle
Agent building, orchestration, and evaluation suites. All integrated. No stitching 5 tools together.

5️⃣ Security & Governance
Role-based, marking-based, purpose-based controls. System-wide branching. Approvals. Checkpoints. What regulated industries ACTUALLY need.

6️⃣ End-to-end Observability
Every tool call. Every data access. Every decision traced. Full model catalog.

Plus 6 more layers: tool services, multimodal compute, development environments, human+AI applications, package/release/deploy, and enterprise automation.

12 layers. One architecture. Production-ready.

The framework wars were the warmup.

The real game is agentic architecture.

Are you still debating agent frameworks? 👇

♻️ Repost for someone building agents who hasn't seen this architecture

Google just gave every AI coding assistant NATIVE knowledge of how to build agents.

15,600 GitHub stars. One command. FREE.

Your coding assistant doesn't know ADK. It hallucinates APIs and invents methods that don't exist.

Here's why this changes everything:

npx skills add google/adk-docs/skills -y

One command. Your coding agent now knows:

1️⃣ ADK Cheatsheet
Full Python API reference. Every method, every parameter. No more hallucinated imports.

2️⃣ ADK Dev Guide
Development lifecycle and coding guidelines. The patterns Google's own team uses.

3️⃣ ADK Deploy Guide
Agent Engine and Cloud Run deployment. Prototype to production in the same workflow.

4️⃣ ADK Eval Guide
Evaluation methodology and scoring. Test your agents before they hit production.

5️⃣ ADK Observability Guide
Tracing, logging, and integrations. See what your agents are ACTUALLY doing.

6️⃣ ADK Scaffold
New agent project in seconds. Right structure from the start.

Works with everything:

⚡ Claude Code: one command install
⚡ Cursor: MCP server config
⚡ Gemini CLI: extension install
⚡ Any MCP-compatible tool

Plus a docs MCP server that connects your editor directly to ADK documentation. Always up to date. Machine-readable.

What this replaces:

Hours of Googling API docs → $0
Debugging hallucinated methods → $0
Manually reading changelogs → $0

ADK has 15,600 stars, 2,800+ dependent projects, and now ships in Python, TypeScript, Go, and Java.

Google isn't just building an agent framework.

They're teaching YOUR coding assistant how to use it.

🔗 Install skills: npx skills add google/adk-docs/skills -y
🔗 Full docs: google.github.io/adk-docs

Are you still letting your coding assistant hallucinate ADK APIs? 👇

♻️ Repost for someone building with Google ADK who doesn't know about this

Google just dropped a FREE agent starter pack.

One command. 60 seconds. Full production stack.

3,100+ GitHub stars. Here's why:

5 agent templates, ready to deploy:

1️⃣ ReAct agent (ADK)
2️⃣ RAG agent: document retrieval + Q&A
3️⃣ Multi-agent with A2A Protocol
4️⃣ Live API: real-time multimodal
5️⃣ LangGraph base

That's the starting point. The infrastructure is what makes it DIFFERENT:

⚡ Cloud Run + Agent Engine deployment
⚡ Terraform configs included
⚡ CI/CD with Cloud Build + GitHub Actions
⚡ Vertex AI evaluation built in
⚡ Interactive playground for testing

One command:

uvx agent-starter-pack create my-awesome-agent

No boilerplate. No config hell. No "works on my machine."

Google gave away what most teams spend weeks building.

For $0.

Which template are you starting with? 👇

♻️ Repost for someone building agents from scratch

The complete AI Agents learning stack for 2026.

31 resources. Every single one free.

Most engineers bookmark 50 resources and finish ZERO.

This isn't a bookmark dump. It's a 5-phase path:

📹 UNDERSTAND

1️⃣ Intro to Large Language Models (Karpathy) → https://lnkd.in/egCPxzEx
2️⃣ Deep Dive into LLMs (Karpathy) → https://lnkd.in/euMCHnyv
3️⃣ Agentic AI Overview (Stanford) → https://lnkd.in/etjh6sYZ
4️⃣ Building and Evaluating Agents → https://lnkd.in/e45S6ZcZ
5️⃣ Building Effective Agents → https://lnkd.in/eKEaupVF
6️⃣ Building Agents with MCP → https://lnkd.in/eVhWN9kK
7️⃣ Building an Agent from Scratch → https://lnkd.in/eY3c5npC

📜 READ THE SCIENCE

1️⃣ ReAct → arxiv.org/abs/2210.03629
2️⃣ Chain-of-Thought Prompting → arxiv.org/abs/2201.11903
3️⃣ Tree of Thoughts → arxiv.org/abs/2305.10601
4️⃣ Reflexion → arxiv.org/abs/2303.11366
5️⃣ Toolformer → arxiv.org/abs/2302.04761
6️⃣ Generative Agents → arxiv.org/abs/2304.03442
7️⃣ RAG Survey → arxiv.org/abs/2312.10997

🗺️ STUDY THE ARCHITECTURES

1️⃣ Google's Agent Whitepaper → https://lnkd.in/ejiEnCdK
2️⃣ Building Effective Agents (Anthropic) → https://lnkd.in/ew-rnBgD
3️⃣ OpenAI's Practical Guide → https://lnkd.in/ef2vzH2C
4️⃣ Claude Code Best Practices → https://lnkd.in/eRBfw9k5
5️⃣ Google's Agent Companion → https://lnkd.in/eezmZrVs

🧑‍🏫 BUILD

1️⃣ HuggingFace Agent Course → https://lnkd.in/gxSFVqhY
2️⃣ MCP with Anthropic → https://lnkd.in/efqmGWaR
3️⃣ Agent Design Patterns → https://lnkd.in/gHhvHAKq
4️⃣ Multi-Agent Systems → https://lnkd.in/gkg3qzMG
5️⃣ Building Browser Agents → https://lnkd.in/etMGU_bi
6️⃣ Computer Use with Anthropic → https://lnkd.in/edR3Qg94
7️⃣ Evaluating AI Agents → https://lnkd.in/eFFiUue5
8️⃣ Building & Evaluating RAG Apps → https://lnkd.in/gWAbB363
9️⃣ Agent Memory → https://lnkd.in/eCf57tqh

🗂️ SHIP

1️⃣ GenAI Agents → https://lnkd.in/gykAzWMt
2️⃣ Microsoft AI Agents for Beginners → https://lnkd.in/efzMR5aU

5 phases. 31 resources. $0.

Bootcamps charge $5,000+ for LESS.

Which phase are you starting with? 👇

♻️ Repost for someone who wants to build agents but doesn't know where to start.

RAG, CoT, ReAct, RLHF, LoRA, MoE

What do these acronyms actually mean?

If you're breaking into AI engineering, these come up constantly. 

Here's the breakdown:

🔹 RAG (Retrieval Augmented Generation)
Instead of relying only on what the model "knows," you retrieve relevant documents and feed them into the prompt. Reduces hallucination. Powers most production AI apps.

🔹 CoT (Chain of Thought)
Prompting the model to "think step by step" before answering. Dramatically improves reasoning on complex tasks. Free performance boost.

🔹 ReAct (Reasoning + Acting)
Agent pattern where the model alternates between thinking and taking actions. Think → Act → Observe → Repeat. Foundation of most agent frameworks.

🔹 RLHF (Reinforcement Learning from Human Feedback)
How ChatGPT became ChatGPT. Train the model to prefer outputs humans rate highly. Why models are helpful instead of just predictive.

🔹 LoRA (Low-Rank Adaptation)
Fine-tune a model by training only a small number of new parameters. Makes fine-tuning accessible without massive GPU costs.

🔹 MoE (Mixture of Experts)
Model architecture where only a subset of parameters activate per token. How Mixtral and Nemotron run 30B+ params on consumer hardware.

Frameworks change every quarter.
These acronyms won't.

💾 Save for your next "wait, what does that stand for?" moment
♻️ Repost so they don't have to ask you.

taplio

The Battle

Harpreet Sahota 🥑

Paolo Perrone

Followers

Posts

Engagements

Likes

Performance ScoreHarpreet Sahota 🥑's LinkedIn Performance

Performance ScorePaolo Perrone's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Harpreet Sahota 🥑

Paolo Perrone

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Harpreet Sahota 🥑's LinkedIn Performance

Performance ScorePaolo Perrone's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreHarpreet Sahota 🥑's LinkedIn Performance