# CU Entrepreneurship Ecosystem Dataset

> 1719 catalogued CU Boulder innovation & entrepreneurship resources from a
> depth-8 saturated crawl of 4,700+ pages plus the CU System I&E impact dashboard
> (filtered to Boulder). Includes 920 tagged people, 123 courses,
> LLM summaries, entities, sequencing edges, and a persona eval set.
> All JSON, CORS-open, no auth. Base: https://cu-entrepreneurship-agent.pages.dev/data/crawl/

## Start here
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/schema.json — machine-readable field docs with ACTUAL enum values and counts. Fetch this first to construct valid filters.
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/API.md — human-oriented guide: schemas, join keys, query recipes.

## Data
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/resources.json — the catalogue (join key: id)
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/relationships.json — link edges between resources
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/enriched.json — LLM summaries + entities per resource
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/entity-graph.json — consolidated people/org/program entities + co-occurrence
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/people.json — faculty/staff/mentors with program affinity scores
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/courses.json — CU Boulder I&E courses + search vocabulary
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/sequence-edges.json — which programs precede/feed into which
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/persona-eval.json — 44 personas, 322 queries, RAG hit rates (weak rows = known gaps)
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/company-journeys.json — 153 historical company journeys: programs touched + outcomes (golden test set)
- https://cu-entrepreneurship-agent.pages.dev/data/crawl/summary.json — crawl stats and depth-saturation proof

## How to query (for LLM agents)
1. Fetch schema.json for valid field values — do not guess enum strings.
2. Filter resources.json client-side (it is a plain array; ~2MB).
   Typical filters: type, audience[], stage[], sectors[], specificity>=6, funding.length>0.
3. Resolve names→people via people.json (roleClass: faculty|staff|mentor|founder-alum).
4. For "what comes next" questions use sequence-edges.json, then stage[] ordering
   (explore→validate→build→launch), then funding amounts ascending.
5. For course questions filter courses.json by level; use searchTerms for synonyms.

## Interfaces
- / — chat navigator (client-side RAG over this dataset)
- /crawl — live crawl dashboard
- /personas — persona & eval explorer
- /people — people & course explorer