External pipelines · RISE project catalogue
DeepResearcher (GAIR-NLP)
An end-to-end RL-trained deep-research agent (arXiv:2504.03160) that learns to plan, retrieve, cross-validate, and self-reflect via reinforcement learning in real-world web environments rather than in simulated retrieval. Ships a 7B HuggingFace checkpoint (DeepResearcher-7b) trained via this pipeline.
Where it sits
left: what it builds on · right: what builds on it · pale: exampleContributed by Shanghai Jiao Tong University GAIR-NLP
How studies reach it
No published study reaches it yet.
Disciplines it reaches
No study reaches it yet.
Solid: published studies. Light: examples.
Computed from the records on this site: what each study, template and specialist names as used, which study extends which, and who contributed what. 0 studies in total.
What it does
Argues that *end-to-end RL on real web environments* — not prompt engineering and not RL on retrieval simulators — is what unlocks emergent cognitive behaviors in research agents (planning, multi- source cross-validation, self-reflection, honest non-answer when evidence is missing). Reports +28.9 points over prompt baselines and +7.2 over RAG-RL baselines.
- Focus
- literature
- Inputs
- research-question
- Outputs
- research-report, citations
- Architecture
- tool-use, iterative-loop, rag-knowledge-base
- Maintained by
- GAIR-NLP, Shanghai Jiao Tong University
- Started
- 2025
Description
Data model- Discipline
- General
- Method family
- Literature review
- Design
- not specified
- Research stage
- Research questionLiterature discoveryLiterature synthesis
- Contributors
- Shanghai Jiao Tong University GAIR-NLP
- Usage
- not used in published research yet
- Source
- RISE project catalogue · projects/landscape · @4c17bae
- Record
- pipeline:deepresearcher · JSON
Solid tags are declared by the source or mapped from its terms; dashed tags are inferred by a published rule. Hover a tag for its provenance.
Bring it into the standard
A pipeline built outside E2ER can meet the standard by describing its steps as a template, attaching the floor of checks and publishing evaluation records. Its authors keep ownership and credit.