From manual scholarship research to a structured platform, with an agentic AI layer on top

We tackled it in two moves. First we turned the manual operation into a structured platform: scholarship records with deadlines, amounts, and status in one place, replacing the spreadsheets, plus essay prep, application tracking, reporting, and parent visibility. With that foundation in place, scholarship discovery was the next bottleneck: the most manual, highest-volume work in the business. So we automated the workflow itself, with a team of nine Claude agents rather than a single assistant: a supervisor that routes, a scout, a writer, a researcher working in a real managed browser, a translator, a student-facing matchmaker, and three sub-agents. Claude models are pinned per agent, with the supervisor and the student-facing matchmaker on separate settings, and each agent's instructions are editable as a setting that hot-reloads in about a minute. One request, like finding scholarships for a new cohort in a geography, now runs the full process: search, extract, enrich, translate, structure, and stage reviewable opportunities for admins. Because the agents act on the platform's own data model, each result lands as a real record with eligibility, deadlines, award amounts, and essay prompts attached, and anything uncertain flagged for human review.