Introducing Rishi (AAR-1): TNSA's First Automated AI Researcher

Rishi (AAR-1) is TNSA's first automated AI researcher, built to accelerate AI progress through continuous research, testing, and iteration.

The goal is not to replace researchers, but to make the research loop faster: more hypotheses explored, more failed ideas filtered early, and more promising directions brought back to human teams.

Rishi can study algorithms, run tests, train models, analyze results, and design follow-up experiments in a continuous cycle of improvement.

Research systems need to work over long horizons, not just single prompts. Rishi is designed for long-running campaigns with restart resilience, checkpointing, and summarization systems that preserve useful context over time.

That structure lets it continue a line of work across experiments while keeping human researchers informed about important decisions and unexpected results.

Rishi works inside a dedicated workspace with files, logs, generated algorithms, and research documents. It also uses TNSA's ARCH-X framework and is connected to the broader direction behind the Tri-Flux Attention paper.

Rishi is an early step toward AI research teams made of autonomous systems working alongside human researchers.

We expect future versions to become better at experiment design, literature synthesis, code generation, benchmark construction, and safely coordinating with other specialized agents.

The most important metric is not whether an automated researcher can produce impressive text. It is whether it can produce useful research progress that survives review, replication, and comparison against strong baselines.