{"work":{"id":"40b60d06-dc1c-4799-b75d-ff1eca653049","openalex_id":"https://openalex.org/W4416411532","doi":"10.48550/arxiv.2507.19457","arxiv_id":"2507.19457","raw_key":null,"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","authors":null,"authors_text":"Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong","year":2025,"venue":"cs.CL","abstract":"Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing one or more LLM prompts, GEPA samples trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems, propose and test prompt updates, and combine complementary lessons from the Pareto frontier of its own attempts. As a result of GEPA's design, it can often turn even just a few rollouts into a large quality gain. Across six tasks, GEPA outperforms GRPO by 6% on average and by up to 20%, while using up to 35x fewer rollouts. GEPA also outperforms the leading prompt optimizer, MIPROv2, by over 10% (e.g., +12% accuracy on AIME-2025), and demonstrates promising results as an inference-time search strategy for code optimization. We release our code at https://github.com/gepa-ai/gepa .","external_url":"https://arxiv.org/abs/2507.19457","cited_by_count":0,"metadata_source":"pith","metadata_fetched_at":"2026-08-05T02:28:24.338817+00:00","pith_arxiv_id":"2507.19457","created_at":"2026-05-09T06:05:33.930128+00:00","updated_at":"2026-08-05T02:28:24.338817+00:00","title_quality_ok":true,"display_title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","render_title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning"},"hub":{"state":{"work_id":"40b60d06-dc1c-4799-b75d-ff1eca653049","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":114,"external_cited_by_count":0,"distinct_field_count":13,"first_pith_cited_at":"2025-09-25T23:25:47+00:00","last_pith_cited_at":"2026-07-08T18:27:40+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-08-23T03:19:16.663329+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":18},{"context_role":"method","n":5},{"context_role":"baseline","n":3},{"context_role":"other","n":1}],"polarity_counts":[{"context_polarity":"background","n":15},{"context_polarity":"use_method","n":5},{"context_polarity":"unclear","n":4},{"context_polarity":"baseline","n":2},{"context_polarity":"contest","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","claims":[{"claim_text":"Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"(bottom) updatesΦ via reflective optimization, additionally consuming the rollout's full text including thoughts, tool calls, errors, and rich feedback (Φc →Φ c+1). MaintainingΦ as a Pareto-frontier population (ratherthanasinglebestprompt)preservesdiversity: differentcontextsspecializetodifferentproblemslices exposing the slow update rule to rich conditioning during training. obtained by improving the textual context under which the model operates [2, 24, 36, 62, 66]. In this work, we introduceF","claim_type":"method","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"open-source Multibody Dynamics Solver, to improve its utility for the agent by, for example, incorporating quaternions and providing more detailed error messages about inconsistent constraints in the assembly and convergence failures. Our approach was inspired by prior work which demonstrated that havingstrong verifiers, i.e. functions which provide robust error / feedback signals to an agentic system [1, 4, 11, 15, 17, 28, 34], can improve performance (perhaps even more than making changes to t","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"General prompting methods (e.g., few-shot [ 3], chain- of-thought [19]) improve reasoning by altering input formats without updating model parameters. To further refine language policies, feedback-driven methods like Reflexion [16] and ExpeL [24] leverage verbal feedback and accumulated experience, while frameworks such as Promptbreeder [6], MIPRO [11], and GEPA [7] treat prompts as optimizable objects via search and validation. Tool use, workflows, and program search.Tool-augmented agents (e.g.","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"evolution baselines and achieves state-of-the-art performance under the same iteration budget. 2 Related Work Agentic Evolution.A growing line of work uses LLMs and agents to iteratively improve artifacts through generation, feedback, and revision [14, 15, 13, 28]. Prompt methods optimize language-model programs or feedback-driven prompts, including DSPy [12], SPO [30], TextGrad [37], and GEPA [1]. Another line studies the automated design and evolution of agentic systems and workflows, such as ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"tasks or 5 call awriter agent(writer), which combines all merged information into a long-form report. time-consuming round of trial-and-error [16], where developers manually tweak prompts and architectures until they reach a reasonable quality threshold [8]. Recently, it has been proposed to shift from manual \"prompt engineering\" towards a more algorithmic \"prompt optimization\" process. Such proposals- including GEPA [1], TextGrad [19], AutoPDL [16], GASO [17], GAPO [5], the Darwin-Gödel Machine","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Blocked: external constraints (missing files, tool failures) prevent progress. The agent must document the specific blocker and any attempted workaround. 4.3 Prompt Optimization via GEPA Rather than relying on a single hand-crafted prompt, each archi- tectural smell type receives its own optimized execution playbook and few-shot demonstrations. We use GEPA [1], a reflective prompt evolution optimizer, to automatically transform human-authored baseline templates into high-quality, smell-specific ","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (18 contexts).","role_counts":[{"n":18,"context_role":"background"},{"n":5,"context_role":"method"},{"n":3,"context_role":"baseline"},{"n":1,"context_role":"other"}]},"error":null,"updated_at":"2026-07-02T21:13:00.031619+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"da421008-c7bc-4087-ae50-fda7d9b97674","orcid":null,"display_name":"Lakshya A Agrawal"},{"id":"b143f9bc-abf9-47bf-8e45-ff560cb0aedc","orcid":null,"display_name":"Shangyin Tan"},{"id":"d726813d-a21a-4cc3-b904-9f10b0802b1d","orcid":null,"display_name":"Dilara Soylu"},{"id":"897ed44a-18b9-489b-908c-eb1a47a983fc","orcid":null,"display_name":"Noah Ziems"},{"id":"49106c85-b65e-403a-8db7-9021bf652711","orcid":null,"display_name":"Rishi Khare"},{"id":"286b1633-271f-4d3a-b1ae-1163151df9f9","orcid":null,"display_name":"Krista Opsahl-Ong"}]},"error":null,"updated_at":"2026-07-02T21:13:00.027934+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T18:19:46.666468+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"AlphaEvolve: A coding agent for scientific and algorithmic discovery","work_id":"76a0f850-d490-4e4f-ab98-8d25df82cd23","shared_citers":16},{"title":"TextGrad: Automatic \"Differentiation\" via Text","work_id":"c52b4841-a4b4-4b4e-bc86-8ef6645a9bbb","shared_citers":10},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":9},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":8},{"title":"Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models","work_id":"c1d09cf9-9176-403d-8028-630bbc0cbc5d","shared_citers":7},{"title":"DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines","work_id":"d490f594-f5fc-47b0-ae6a-6550e50fe095","shared_citers":7},{"title":"arXiv preprint arXiv:2507.21046 , year=","work_id":"f5de9511-98bf-411c-9eba-e8b7a914ec18","shared_citers":6},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":6},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":5},{"title":"isolation level","work_id":"c0d964a0-02a1-4ee2-a596-028489ba753f","shared_citers":5},{"title":"Le, Denny Zhou, and Xinyun Chen","work_id":"cacd55e1-79db-4dd6-9f1b-794fc09d75bb","shared_citers":5},{"title":"Memevolve: Meta-evolution of agent mem- ory systems","work_id":"69fd3342-cd0c-43d2-b12c-2d266dc000b2","shared_citers":5},{"title":"Promptbreeder: Self-referential self-improvement via prompt evolution","work_id":"01508be0-d975-4ff6-8045-1e0b229fbab4","shared_citers":5},{"title":"Reflexion: Language Agents with Verbal Reinforcement Learning","work_id":"778f739e-5f55-4961-8a2a-e4736a2757f4","shared_citers":5},{"title":"Thetaevolve: Test-time learning on open problems","work_id":"0c9ee4af-c6aa-47e7-b1cc-863823529ecb","shared_citers":5},{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":5},{"title":"Adae- volve: Adaptive llm driven zeroth-order optimization","work_id":"284cc2bd-522c-4674-99da-c07e54d2afff","shared_citers":4},{"title":"Chrisantha Fernando, Dylan Sunil Banarse, Henryk Michalewski, Simon Osindero, and Tim Rocktäschel","work_id":"10037f9e-62d9-4dd1-a9c1-44be77032524","shared_citers":4},{"title":"Codeevolve: An open source evolutionary coding agent for algorithm discovery and optimization","work_id":"b31e9fdb-cb9e-45bb-8e54-e6fc74ca613a","shared_citers":4},{"title":"Large language models are human-level prompt engineers","work_id":"29231305-69fd-45ff-b0cd-4dc3bd9d798e","shared_citers":4},{"title":"Learning to discover at test time","work_id":"e6d32375-24bd-4c3a-a7b8-fe20121f31b1","shared_citers":4},{"title":"OpenHands: An Open Platform for AI Software Developers as Generalist Agents","work_id":"f1762ea0-e382-4f38-a28c-adc643789859","shared_citers":4},{"title":"Shinkaevolve: Towards open-ended and sample-efficient program evolution","work_id":"41f30c7f-ee82-49b7-9ce9-463d3124269c","shared_citers":4},{"title":"SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering","work_id":"01826cd9-a652-403c-a2ec-531da9fe2b6a","shared_citers":4}],"time_series":[{"n":36,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T18:19:40.062643+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T18:19:29.711302+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","claims":[{"claim_text":"Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"(bottom) updatesΦ via reflective optimization, additionally consuming the rollout's full text including thoughts, tool calls, errors, and rich feedback (Φc →Φ c+1). MaintainingΦ as a Pareto-frontier population (ratherthanasinglebestprompt)preservesdiversity: differentcontextsspecializetodifferentproblemslices exposing the slow update rule to rich conditioning during training. obtained by improving the textual context under which the model operates [2, 24, 36, 62, 66]. In this work, we introduceF","claim_type":"method","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"open-source Multibody Dynamics Solver, to improve its utility for the agent by, for example, incorporating quaternions and providing more detailed error messages about inconsistent constraints in the assembly and convergence failures. Our approach was inspired by prior work which demonstrated that havingstrong verifiers, i.e. functions which provide robust error / feedback signals to an agentic system [1, 4, 11, 15, 17, 28, 34], can improve performance (perhaps even more than making changes to t","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"General prompting methods (e.g., few-shot [ 3], chain- of-thought [19]) improve reasoning by altering input formats without updating model parameters. To further refine language policies, feedback-driven methods like Reflexion [16] and ExpeL [24] leverage verbal feedback and accumulated experience, while frameworks such as Promptbreeder [6], MIPRO [11], and GEPA [7] treat prompts as optimizable objects via search and validation. Tool use, workflows, and program search.Tool-augmented agents (e.g.","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"evolution baselines and achieves state-of-the-art performance under the same iteration budget. 2 Related Work Agentic Evolution.A growing line of work uses LLMs and agents to iteratively improve artifacts through generation, feedback, and revision [14, 15, 13, 28]. Prompt methods optimize language-model programs or feedback-driven prompts, including DSPy [12], SPO [30], TextGrad [37], and GEPA [1]. Another line studies the automated design and evolution of agentic systems and workflows, such as ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"tasks or 5 call awriter agent(writer), which combines all merged information into a long-form report. time-consuming round of trial-and-error [16], where developers manually tweak prompts and architectures until they reach a reasonable quality threshold [8]. Recently, it has been proposed to shift from manual \"prompt engineering\" towards a more algorithmic \"prompt optimization\" process. Such proposals- including GEPA [1], TextGrad [19], AutoPDL [16], GASO [17], GAPO [5], the Darwin-Gödel Machine","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Blocked: external constraints (missing files, tool failures) prevent progress. The agent must document the specific blocker and any attempted workaround. 4.3 Prompt Optimization via GEPA Rather than relying on a single hand-crafted prompt, each archi- tectural smell type receives its own optimized execution playbook and few-shot demonstrations. We use GEPA [1], a reflective prompt evolution optimizer, to automatically transform human-authored baseline templates into high-quality, smell-specific ","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (18 contexts).","role_counts":[{"n":18,"context_role":"background"},{"n":5,"context_role":"method"},{"n":3,"context_role":"baseline"},{"n":1,"context_role":"other"}]},"error":null,"updated_at":"2026-07-02T21:13:00.034169+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","claims":[{"claim_text":"Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T18:20:19.189837+00:00"}},"summary":{"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","claims":[{"claim_text":"Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"AlphaEvolve: A coding agent for scientific and algorithmic discovery","work_id":"76a0f850-d490-4e4f-ab98-8d25df82cd23","shared_citers":16},{"title":"TextGrad: Automatic \"Differentiation\" via Text","work_id":"c52b4841-a4b4-4b4e-bc86-8ef6645a9bbb","shared_citers":10},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":9},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":8},{"title":"Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models","work_id":"c1d09cf9-9176-403d-8028-630bbc0cbc5d","shared_citers":7},{"title":"DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines","work_id":"d490f594-f5fc-47b0-ae6a-6550e50fe095","shared_citers":7},{"title":"arXiv preprint arXiv:2507.21046 , year=","work_id":"f5de9511-98bf-411c-9eba-e8b7a914ec18","shared_citers":6},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":6},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":5},{"title":"isolation level","work_id":"c0d964a0-02a1-4ee2-a596-028489ba753f","shared_citers":5},{"title":"Le, Denny Zhou, and Xinyun Chen","work_id":"cacd55e1-79db-4dd6-9f1b-794fc09d75bb","shared_citers":5},{"title":"Memevolve: Meta-evolution of agent mem- ory systems","work_id":"69fd3342-cd0c-43d2-b12c-2d266dc000b2","shared_citers":5},{"title":"Promptbreeder: Self-referential self-improvement via prompt evolution","work_id":"01508be0-d975-4ff6-8045-1e0b229fbab4","shared_citers":5},{"title":"Reflexion: Language Agents with Verbal Reinforcement Learning","work_id":"778f739e-5f55-4961-8a2a-e4736a2757f4","shared_citers":5},{"title":"Thetaevolve: Test-time learning on open problems","work_id":"0c9ee4af-c6aa-47e7-b1cc-863823529ecb","shared_citers":5},{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":5},{"title":"Adae- volve: Adaptive llm driven zeroth-order optimization","work_id":"284cc2bd-522c-4674-99da-c07e54d2afff","shared_citers":4},{"title":"Chrisantha Fernando, Dylan Sunil Banarse, Henryk Michalewski, Simon Osindero, and Tim Rocktäschel","work_id":"10037f9e-62d9-4dd1-a9c1-44be77032524","shared_citers":4},{"title":"Codeevolve: An open source evolutionary coding agent for algorithm discovery and optimization","work_id":"b31e9fdb-cb9e-45bb-8e54-e6fc74ca613a","shared_citers":4},{"title":"Large language models are human-level prompt engineers","work_id":"29231305-69fd-45ff-b0cd-4dc3bd9d798e","shared_citers":4},{"title":"Learning to discover at test time","work_id":"e6d32375-24bd-4c3a-a7b8-fe20121f31b1","shared_citers":4},{"title":"OpenHands: An Open Platform for AI Software Developers as Generalist Agents","work_id":"f1762ea0-e382-4f38-a28c-adc643789859","shared_citers":4},{"title":"Shinkaevolve: Towards open-ended and sample-efficient program evolution","work_id":"41f30c7f-ee82-49b7-9ce9-463d3124269c","shared_citers":4},{"title":"SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering","work_id":"01826cd9-a652-403c-a2ec-531da9fe2b6a","shared_citers":4}],"time_series":[{"n":36,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"d726813d-a21a-4cc3-b904-9f10b0802b1d","orcid":null,"display_name":"Dilara Soylu","source":"manual","import_confidence":0.72},{"id":"286b1633-271f-4d3a-b1ae-1163151df9f9","orcid":null,"display_name":"Krista Opsahl-Ong","source":"manual","import_confidence":0.72},{"id":"da421008-c7bc-4087-ae50-fda7d9b97674","orcid":null,"display_name":"Lakshya A Agrawal","source":"manual","import_confidence":0.72},{"id":"897ed44a-18b9-489b-908c-eb1a47a983fc","orcid":null,"display_name":"Noah Ziems","source":"manual","import_confidence":0.72},{"id":"49106c85-b65e-403a-8db7-9021bf652711","orcid":null,"display_name":"Rishi Khare","source":"manual","import_confidence":0.72},{"id":"b143f9bc-abf9-47bf-8e45-ff560cb0aedc","orcid":null,"display_name":"Shangyin Tan","source":"manual","import_confidence":0.72}]}}