The work centred on a common delivery problem: project knowledge was spread across too many formats, and too much of the analysis depended on people manually reopening files, retracing evidence, and rebuilding the same context from scratch. A useful answer was not just a search result. It needed to sit inside a project structure, belong to a specific area of analysis, and remain traceable enough to be reviewed later.
I designed the workflow so document processing could happen in the background, without blocking the user while larger uploads were being prepared for analysis. Once ingested, source material could be broken into searchable units, while tabular content was handled with more care. Spreadsheet-heavy questions needed a different path from narrative material because sales tables, category breakdowns, and time-series exports rarely behave like reports, slides, or written documents.
The platform also needed a practical review layer. Answers were versioned, review states were built in, and only approved material flowed into summaries and synthesis. That matters in real research work, where teams often need to test a question, re-run it, compare drafts, and decide what is reliable enough to carry forward.
The result is a backend that supports day-to-day research operations with less manual rework. Analysts can work through large sets of source material in a more structured way, reviewers can check what should progress into summary outputs, and reporting becomes an extension of the analysis process rather than a separate cleanup exercise.