The Asari AI blog
Experiments, breakthroughs, and the systems behind them.Same Silicon, More Intelligence for Automated Driving
Asari AI’s co-inventor agents sped up an automaker’s production behavior prediction model by up to 14.6x, recovering compute budget for more intelligence on the same silicon.
Speeding up end-to-end inference with self-improving agents
Early results on how our co-inventor agents improve the full AI inference stack using a rigorous self-improving process.
What does it take for AI to build space software?
Building software for space leaves zero margin for error. We explore what it took for our AI agents to migrate the space-grade math library libmcs from C to Safe Rust, preserving bit-level precision guarantees and catching subtle issues that even expert reviewers missed.
Migrating critical systems to Safe Rust with reliable agents
Our AI co-inventor successfully translated critical C libraries to memory-safe Rust, demonstrating a path forward for building AI systems that can tackle complex, high-stakes engineering work as true collaborators.
Enabling intelligent search for AI agents
We introduce EnCompass, a new framework for LLM-based agents that cleanly separates workflow design from inference-time strategy. By compiling agent workflows into a structured search space, our framework makes it easy to experiment with strategies like tree search and improve reliability with minimal code changes.