Talk to your documents instead of reading all of them
Who it is for: Researchers and analysts
The problem today
Section titled “The problem today”A hundred PDFs, half of them scans with no text layer, and finding one clause means Ctrl+F across all of them. Half the answers are trusted rather than checked.
How Erebros handles it
Section titled “How Erebros handles it”Documents are read, the mess left by OCR is repaired, and the result is indexed. Answers come back with the page they were taken from, so a claim can be checked instead of trusted, and nothing leaves the machine the files already sit on.
The pipeline
Section titled “The pipeline”- D-Forge: Read the files
- MinerU: OCR the scans
- N-Forge: Index the sources
- N-Forge: Ask with citations
- Memory: Keep the answer