In this tutorial, we build a compact, efficient framework that demonstrates how to convert tool documentation into standardized, callable interfaces, register these tools in a central system, and execute them as part of an...
How to convert complex multilingual documents (dense layouts, small scripts, formulas, diagrams, and handwriting) into faithfully structured Markdown/JSON with state-of-the-art accuracy while keeping inference latency and memory low enough to be suitable for real-world...
A research team from Google Research, Google DeepMind, and Yale University released C2S-Scale 27Ba basic model for single-cell analysis containing 27 billion parameters Gemma-2. This model formalizes single-cell RNA-seq (scRNA-seq) profiles as “Cell Sentence”–...
Agent systems are stochastic, context-sensitive, and policy-bound. Traditional quality assurance (unit tests, static prompts, or scalar “LL.M. as judge” scores) fails to expose multiple rounds of vulnerabilities and provide a weak audit trail. Developer...
In this tutorial, we build an AI-driven cryptographic proxy system that combines the benefits of classic cryptography with adaptive intelligence. We designed the agent to perform hybrid encryption using RSA and AES, generate digital...
Agent systems are stochastic, context-sensitive, and policy-bound. Traditional quality assurance (unit tests, static prompts, or scalar “LL.M. as judge” scores) fails to expose multiple rounds of vulnerabilities and provide a weak audit trail. Developer...
What would you build if you could run reinforcement learning (RL) post-training on a 32B LLM (on a single H100) in 4-bit NVFP4 with BF16-level accuracy and 1.2–1.5× step speedup? NVIDIA researchers (with collaborators...
In this tutorial, we’ll explore how to build a context-collapsed LLM agent that efficiently solves long-term, complex tasks by intelligently managing limited context. We design agents to break large tasks into smaller subtasks, perform...
Anthropic release Claude Haiku 4.5a latency-optimized “small” model that can provide Encoding performance levels similar to Claude Sonnet 4 while running More than twice the speed, one-third less cost. The model is immediately available...
How would your agent stack change if a policy could be trained purely on its outcome-based deployment (no rewards, no demonstrations) but beat imitation learning on eight benchmarks? Yuan Super Intelligence Laboratory proposed “early...