Bootstrapping Conversational Recommendation Agents At Spotify: Synthetic Data Generation and Self-Improvement Loops

arXiv:2609.30297 · cs.CL, cs.AI, cs.IR · Submitted 2026-09-16 · Read on arXiv

cs.CL, cs.AI, cs.IR

Submitted: 2026-09-16

Updated: 2026-09-16

Code: https://github.com/karpathy/autoresearch

Project page: https://gepa-ai.github.io

Terminology

• Content Refinement — 8× in this paper
• LLM-as-a-Judge — 8× in this paper · explained in 3 episodes
• Trace — 7× in this paper · explained in 2 episodes
• Failure Modes — 6× in this paper · explained in 3 episodes
• Synthetic Data Generation — 6× in this paper · explained in 5 episodes
• Benchmark — 5× in this paper · explained in 2 episodes
• Intent Shift — 5× in this paper
• Iterative Refinement — 5× in this paper · explained in 3 episodes
• Adaptation — 4× in this paper · explained in 1 episode
• Instruction Retention — 4× in this paper
• Anaphora Resolution — 3× in this paper
• Agent Harness — 2× in this paper · explained in 2 episodes
• Agentic Search — 2× in this paper · explained in 1 episode
• Alignment — 2× in this paper · explained in 1 episode
• Grounding — 2× in this paper · explained in 3 episodes
• Prompt Engineering — 2× in this paper · explained in 4 episodes
• Separate Input — 2× in this paper
• Accuracy — 1× in this paper · explained in 1 episode
• Agentic Approaches — 1× in this paper · explained in 1 episode
• Consistency — 1× in this paper · explained in 3 episodes
• Evaluation Dimensions — 1× in this paper · explained in 3 episodes
• Hallucination — 1× in this paper · explained in 3 episodes
• Large Language Models (LLMs) — 1× in this paper · explained in 6 episodes
• Personalization — 1× in this paper · explained in 2 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Synthetic Data Generation Pipeline — 1× in this paper · explained in 2 episodes
• Transformer — 1× in this paper · explained in 2 episodes

Sources

• GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
• Evaluating Large Language Models Trained on Code
• TalkPlayData 2: An Agentic Synthetic Data Pipeline for Multimodal Conversational Music Recommendation
• TalkPlay-Tools: Conversational Music Recommendation with LLM Tool Calling
• ACON: Optimizing Context Compression for Long-horizon LLM Agents
• LLMs Get Lost In Multi-Turn Conversation
• Meta-Harness: End-to-End Optimization of Model Harnesses
• Talk the Walk: Synthetic Data Generation for Conversational Music Recommendation
• Text2Tracks: Prompt-based Music Recommendation via Generative Retrieval
• LLMJudge: LLMs for Relevance Judgments
• $\tau$-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains
• TextGrad: Automatic "Differentiation" via Text
• Conversational Information Seeking
• Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering