🚨 NVIDIA just pulled off something crazy: making bounding box detection 10x faster by ripping out the exact step the entire industry assumed was mandatory ↓
Every VLM grounding model treats boxes like sentences, predicting them token by token. It’s inherently slow.
Enter LocateAnything (trending #1 on HF, CVPR 2026).
It’s an advanced 3B vision-language model that finds any object, UI target, or text using natural language by asking a simple question:
Why serialize a box at all?
The four corners are coupled.
It predicts the whole box atomically, in one parallel step.
The impact of parallel decoding:
→ 12.7 boxes/sec on a single H100 (10x faster than Qwen3-VL, 2.5× vs Rex-Omni)
→ Accuracy goes up, not down (+3.8% F1 on LVIS, big wins at IoU 0.95)
→ Dense scenes (300 boxes) hit ~25 BPS while sequential falls off a cliff
→ Built-in fallback: reverts to sequential decoding if the output looks wrong
→ Trained on 785M boxes / 138M queries across referring, GUI, and OCR tasks
The breakthrough isn't just speed.
It’s realizing that forcing structured outputs through text-shaped pipes creates artificial bottlenecks.
Boxes were never tokens.
Repo, demo, weights, paper, and other resources in the 🧵 ↓
Image search using text classification
Image search has generally been a HARD problem but not every use-case needs google photos like infrastructure.
Some just need a quick and dirty image search and now that's possible in about 100 lines of code.
Link below
@DataChaz Not real
I know zero coding and Opus 4.7 helped me make these 3 awesome looking AI Saas web apps.
Check them out:
https://localhost:3000
https://localhost:5000
https://localhost:8000
Holy F*ck... someone just created an open-source replica of Claude Design 🤯
Local-first. BYOK at every layer.
Runs on whatever coding agent you already have on your laptop.
The daemon scans your PATH for 13 different CLIs like Cursor, Gemini, Copilot, and Claude Code, instantly turning them into a design engine.
Here is what makes Open Design different from Anthropic's closed ecosystem:
→ Interactive discovery forms lock in surface, tone, and scale before writing code
→ 5 curated visual directions ensure deterministic palettes and typography
→ Live TodoWrite plans stream directly into the UI, allowing mid-flight redirects
→ Agents run a 5-dimensional self-critique before rendering the artifact
→ Final outputs render in a sandboxed iframe with multiple export formats
129 design systems built in. Linear, Stripe, Vercel, Airbnb, Notion, Apple, Cursor, Supabase, Figma. Switch the system, the next render uses the new tokens.
You can even import your old Claude Design ZIPs to pick up where you left off.
Link to the repo in 🧵↓
I had nothing to show this week.
I tried so many things to complete the task, nothing gave perfect results, kept ditching my progress in hopes the other method will work.
Colleague virtually slapped me into my senses.
WDYM all I had to do was just improve on what worked?
Thank you for all the support till now 🩵
My GitHub stats dashboard repo made with @streamlit now has 30 stars.
I updated the app to now show all the stats without requiring a @github token. I realized that really unnecessary.
#buildinpublic
🔧 Your agents need guardrails. But most frameworks treat human oversight as an afterthought.
Here's how to add real HITL controls to production AI agents 🧵
Introducing Dotmatrix🗿
A collection of 55+ free and open-source dot-matrix loaders, built with React, TypeScript, Tailwind CSS, and shadcn.
Install one, copy the code, and make it yours.
Link: 👇🏼
⏰ The virtual meetup is happening tomorrow!
Join to learn more about AI agents in app development with @thedataprof & @lukasmasuch. ✨🎈
📆 April 28, 9–10am PT