Automated API JSON parsing for SAP ERP intelligence
Automated API JSON parsing for SAP ERP intelligence converts each JSON payload attached to a record into searchable passages, so the assistant answers from the file rather than from fields someone typed.
How does automated API JSON parsing feed SAP ERP intelligence?
Automated API JSON parsing for SAP ERP intelligence converts each attached JSON payload into passages when the file is saved, so the assistant can cite the file instead of guessing from record fields.
Where the JSON payload sits in SAP
SAP keeps the file on the business object through Generic Object Services or ArchiveLink, beside the order, invoice, or vendor. The typed fields on that record are already searchable. The JSON payload is not, until automated API JSON parsing copies its words into passages keyed to the same record. ERP intelligence that only reads the fields is intelligence about what a person typed, not about what the file says.
What a JSON file actually yields
JSON is text, but a raw blob of braces retrieves poorly. The parser flattens each value to a path, such as customers.0.name: Acme, so the words that say what the value is travel with the value. Pretty-printed punctuation is not what gets embedded. A payload that is not UTF-8 fails rather than being indexed as replacement characters.
The API call that feeds SAP
An integration flow or an attachment event can POST those bytes when the document is linked. The response is JSONL, one passage per line, with the record id in metadata for the retrieval store beside SAP. Asking a model to reread the JSON payload inside chat for every question never builds that index. Automated API JSON parsing runs when the file is saved, once, and later questions retrieve.
What the answer is allowed to cite
After a clean parse, an answer should cite the passage and the JSON payload it came from. A file that yields no text is a skip, named as a skip, not a blank passage that crowds real hits. The parsers are Remade with Rust crates, and RAG Converter is part of MATA. The longer note on this format is RAG for structured data.