MVP: Define structured recommendation output schema

[andjump]

Context

See:

• docs/architecture.md §6 (Implementation notes)
• docs/PRD.md §8 (MVP scope)

LLM output must be predictable and UI-safe.

Task

Define a structured response schema for Talent Scout recommendations.

Schema should support:

• max 3–5 opportunities
• per-opportunity:
  • title
  • plain-language summary
  • why-this-fits explanation
  • eligibility notes (if any)
  • next-step CTA

Acceptance criteria

• Output is machine-parseable (e.g. JSON)
• UI does not rely on free-form text blobs
• Over-long responses are rejected or trimmed safely

Out of scope

• Free-form chat transcript rendering