AI model training
Build scholarly training corpora without crawling the open web.
Use fulltext, PDFs, source manifests, and provider rankings to assemble domain-specific datasets for fine-tuning, evaluation, and retrieval pipelines.
How do you keep a model corpus current without building a crawler first?
Search or list only artifact-backed works for your topic.
Fetch fulltext in bulk and keep PDFs for inspection or multimodal extraction.
Store manifest data so every training item keeps source and license context.
Implementation pattern
A practical corpus workflow
Discover
Use search for relevance or list for exhaustive ingestion. Either path stays inside the content-backed corpus.
Retrieve
Fetch fulltext, PDFs, or manifests only for the records your workflow accepts.
Trace
Keep provider, source URL, checksum, and rank beside every downstream artifact.
Example
Start with one request
curl "https://scholar.citedevidence.com/v1/research-corpus/search?q=materials+science+solid+electrolyte&has_fulltext=true&limit=50" -H "X-API-Key: ce_xxxxxxxxx"
Then pull the artifact
Once your workflow accepts a work ID, call /text, /texts,/pdf, or /manifest depending on what your client needs next.
Build with CitedEvidence.
Discover literature with Scholar, retrieve available paper content with CEDOA, and connect research tools through MCP.