🚨EL WEB SCRAPING HA MUERTO🚨
Han creado PixelRAG.
Un sistema open source que pasa de parsear HTML.
En vez de convertir una web en texto...
hace una captura de pantalla.
Y luego un modelo vision-language lee la respuesta directamente desde los píxeles.
Brutal.
Porque el scraping tradicional rompe muchísima información:
→ tablas
→ gráficos
→ layouts
→ columnas
→ botones
→ PDFs
→ páginas mal estructuradas
Todo eso se pierde cuando conviertes una web en texto plano.
PixelRAG hace algo mucho más simple:
indexa la página tal y como la ve una persona.
El equipo ha creado un índice visual de Wikipedia entero:
+30 millones de screenshots.
Y aun así supera al mejor RAG basado en texto por un 18,1% en QA.
Tiene también un plugin para Claude Code.
Básicamente le da ojos a Claude.
Le pasas una URL, un paper de arXiv o tu web local…
y Claude no scrapea el DOM.
La mira.
Cómo funciona:
→ renderiza webs, PDFs e imágenes en tiles
→ los embebe con Qwen3-VL-Embedding
→ construye un índice FAISS
→ busca visualmente sobre píxeles
Todo open source bajo Apache-2.0.
Link al repo en los comentarios 👇🏻
Andrew Ng just dropped a 3-hour course on how to become an AI Engineer in 2026:
• 00:00 - How to build agentic AI systems
• 04:25 - Future of AI engineering
• 23:38 - AI Prompting full course
• 2:52:17 - Creating an app with AI in 30 minutes
This 3-hour watch could replace 10 AI engineering courses on the internet.
Watch it today, then read how to run a self-improving system in the article below.
https://t.co/cZAtJnghcx
DeerFlow (Deep Exploration and Efficient Research Flow) is an open-source super agent harness that orchestrates sub-agents, memory, and sandboxes to do almost anything — powered by extensible skills.
One image + text + camera trajectory = controllable worlds. All on a single GPU.
Our research team just released SANA-WM, a 2.6B open source world model natively trained for 60-second video generation with precise camera control.
@hthieblot The hard part isn’t working hard once.
It’s staying consistent when nobody is watching, and the results are still invisible.
That’s where startups are built.
The repo has 10 agencies, each with a specific job title: Sales Agent, Market Researcher, Earnings Assessor, GL Matcher, End-of-Month Registerer, KYC Screener… Each agency builds a sales pitch, writes a professional overview, matches the data, closes the end-of-month register.
Anthropic just relaxed a repo, 76 days and 18 stars. It's called Claude for Financial Services — and it's not a direct-to-market product.
https://t.co/YoAoLVI5ub
@hthieblot I understand that feeling because I’ve experienced it too, and I know how hard that period can be. Your posts have motivated me many times to keep growing and improving myself. Thank you for that.
- AI needs more than data.
- It needs a body.
At HealAI, we’re building compact hardware that captures real respiratory signals, so AI can assist doctors with better lung insights.
@gethealai
We’re very excited to have been selected for Canopy and would love to join.
So excited to have been selected for Canopy and would love to join
@gethealai, @hthieblot, @fdotinc
Urban Traffic Intelligence via Computer Vision
Detection & Tracking: Multi-class vehicle recognition with real-time trajectory tracking (ByteTrack/BoT-SORT).
Zone Analytics: Automated entry/exit counting and illegal parking detection via Polygon Zones.
Introducing Claude Managed Agents: everything you need to build and deploy agents at scale.
It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days.
Now in public beta on the Claude Platform.
i read through the entire claude code source code so u dont have to
11 layers of architecture. 60+ tools. 5 compaction strategies. subagents that share prompt cache.
most people are using maybe 10% of what this thing can do.
heres everything i found: