The need
Problem
A shared delivery model increased the need to preserve and update customer context. That information was spread across a large collection of Word documents.
The build
Engineering approach
I built a Python ingestion process for more than 700 DOCX files and stored extracted information in a relational database. A web UI provided conventional search and navigation alongside an AI-assisted conversational interface. The ingestion process also identified documents it could not parse so they could be reviewed.
The result
Outcome
The prototype demonstrated that a large collection of unstructured documents could be converted into structured, searchable context while also supporting conversational retrieval. Failed ingestion cases remained visible for human review rather than being silently discarded.
Looking back
Lessons & growth
A retrieval interface depends on reliable ingestion as well as search. Identifying documents that could not be parsed was part of making the captured information useful.