Distilabel is a framework for synthetic data and AI feedback for engineers who need fast, reliable and scalable pipelines based on verified research papers.

A continually updated list of literature on Reinforcement Learning from AI Feedback (RLAIF)

Official Implementation of VideoDPO

code for paper Query-Dependent Prompt Evaluation and Optimization with Offline Inverse Reinforcement Learning

ZYN: Zero-Shot Reward Models with Yes-No Questions

Synthetic data for fine tuning LLM

Code and data for "Timo: Towards Better Temporal Reasoning for Language Models" (COLM 2024)

A curated and updated list of relevant articles and repositories on Reinforcement Learning from AI Feedback (RLAIF)

distilled Self-Critique refines the outputs of a LLM with only synthetic data

An implementation of a vanilla RLAIF pipeline, utilizing GPT-2-Large for the summarization task with the TL;DR dataset.

Production-ready RLAIF trading system with multi-agent Claude AI that learns from market outcomes. Features 60+ indicators, foundation models, and serverless deployment.

RewardAnything: Generalizable Principle-Following Reward Models

Code for the paper "Improving Socratic Question Generation using Data Augmentation and Preference Optimization"

RankPO: Rank Preference Optimization

🧠 Enhance AI conversations with Cognio, a persistent memory server that retains context and enables meaningful semantic search across sessions.

(Stepwise controlled Understanding for Trajectories) -- “agent that learns to hunt"

RLAF: Reinforcement Learning from Agentic Feedback - A unified framework for training AI agents with multi-perspective critic ensembles

Production-grade CLI coding agent built from scratch — ReAct loop, 11 tools, RLAIF scoring via Grok 4, cross-session memory, JSONL tracing. 7.5× faster than LangChain baseline.

Experimental graph-centric architecture for concept learning and idea synthesis with LLM teachers, evolutionary graph blending, and preference feedback.

verification as a service: moving beyond llm as a judge