RL-Trained Recursive Language Models
The first models trained with reinforcement learning to develop autonomous recursive decomposition and cost-aware behavior — shipped as lightweight LoRA adapters that run on commodity hardware.
We build AI systems that accelerate scientific discovery — not by replacing the scientist, but by removing the architectural ceilings that prevent any single mind from holding an entire field at once.
Science now produces more knowledge than any mind — human or artificial — can fully hold. The frontier is not intelligence. It is memory. We are building the first AI systems that scale with the world's knowledge rather than being bounded by it.
MSc in Computer Science, ex AWS, now a Visiting Researcher at ETH Zurich’s Agentic Systems Lab. I focus on post-training for language models — especially reinforcement learning and agentic systems — and apply that work at Anadromi Labs.
Today’s literature-review AI systems are limited by fixed context windows: beyond a point, they truncate evidence or lose nuance.
We train Recursive Language Models with reinforcement learning so they can decompose large corpora, preserve full evidence, and optimize for quality and compute cost at scale.
Read the full research article →The first models trained with reinforcement learning to develop autonomous recursive decomposition and cost-aware behavior — shipped as lightweight LoRA adapters that run on commodity hardware.
Our first product: a system that can process an entire research field — millions of papers — without truncation, without sampling, without information loss. Auditable, reproducible, and improving with every run.
We're working with research institutions, clinical teams, and life sciences organisations to deploy AI-powered systematic review. If your work is bottlenecked by the scale of existing literature, we want to hear from you.