Case study · 03
Kataba
Retrieval for geoscience and other technical documents. People searching legacy reports, scans, and well files need the sentence that answers the question, and they need to open the page it came from.
What was built
- Document processing, including scans and legacy files
- Retrieval over the processed material
- Room for a question asked in the language of the domain
- An answer linked to the document, the page, and the passage
What a cited answer has to contain
- Question Asked against a named set of documents.
- Passage The words in the source that the answer relies on.
- Citation Document, page, and passage, attached to the sentence that used them.
Linked evidence
The answer stays attached to the document, the page, and the passage it came from.
Decision
An answer stays linked to the document, the page, and the passage it came from.
What shipped
A public site that describes document processing, retrieval, and source-cited answers.
kataba.ai (opens in a new tab) Vlad’s role, on his portfolio