docs: update the organization card

README.md

@@ -7,10 +7,10 @@ sdk: static
 pinned: false
 ---
 
-AutoTrust AI Lab is a frontier lab of 10 researchers building large-scale, long-horizon agentic models — and the training and inference systems to run them.
+AutoTrust AI Lab is an applied AI neolab of 10 researchers building large-scale, long-horizon agentic models — and the training and inference systems to run them.
 
-We're exploring recursive self-improvement: how AI learns from real experience, and how diverse model populations collaborate to elevate overall system intelligence. Our agents are built to compress the scientific research loop and accelerate discovery at scale.
+Our core edge is a unique post-training recipe for open-source foundation models, built on three pillars: Blocks of Experts architectures, quantization, and model pruning — compressing frontier capability into efficient, deployable systems. We are avid supporters of NVIDIA DGX Spark, and will release pruned and quantized versions of frontier open-source models purpose-built for DGX Spark environments and use cases.
 
-[PaperGuru](https://paperguru.ai) is our flagship AI research agent for serious academic work — handling the full workflow from literature synthesis and methodology through experiment execution, analysis, and writing to top-venue standard. It scored 66.05% on OpenAI's PaperBench, a benchmark for autonomous reproduction of cutting-edge ML research, and has already contributed to 20+ acceptances at peer-reviewed top-tier venues. A stronger result — 96% on PaperBench — is being finalized ahead of release.
+We're exploring recursive self-improvement: how AI learns from real experience, and how diverse model populations collaborate to elevate overall system intelligence. Our agents are built to compress the scientific research loop and accelerate discovery at scale.
 
-ScienceGuru — our AI-native platform for the full scientific process, launching July 2026.
+ScienceGuru — our AI-native platform for the full scientific process — is available on macOS and Windows desktop apps and the web, with an Ubuntu Pro app (supporting DGX OS) coming soon.