grants/AI_STUDIO_PLAN.md

AI Studio — Implementation Plan (Level-ups 1, 2, 3)

Extends the existing AI proxy pattern: src/types/ai.ts → server/routes/generate.ts (schema + prompt + mock + dispatch) → src/lib/ai.ts → component.

All three are pre-fillable from our Workspace profile (Degelas → RWA) and bilingual-aware (we pass locale so output language matches the UI).

---

1. AI Grant Application Studio (feature: "grant_studio")

A full, editable, exportable multi-section application — beyond the quick drafter.

Input: { grantId, grantName, founder, locale } Output sections:

• executiveSummary
• problemStatement
• solutionDescription
• innovationStatement
• technicalApproach (TRL progression)
• workPackages[] (name, focus, months, deliverable)
• consortium[] (partner, role, country)
• budgetBreakdown[] (category, share, justification)
• impactKpis[] (metric, target, timeframe)
• risks[] (risk, mitigation)
• timeline[] (milestone, month)
• complianceChecklist[] (requirement, met)

UI: accordion of editable sections + "Download full application" (.txt/.md).

2. Partner Outreach Generator (feature: "partner_outreach")

Generates a complete outreach kit in the chosen language.

Input: { partnerName, partnerType, ask, founder, locale } Output:

• subject
• emailBody
• linkedinMessage
• followUpEmail
• meetingAgenda[]
• talkingPoints[]
• mouOutline[] (section, content)

UI: partner picker (UM6P, OCP, GEFF bank, EU utility, …) + copy-to-clipboard per block.

3. Financial Model Generator (feature: "financial_model")

Builds a 3-year model + grant/blended-finance stack for the RWA layer.

Input: { businessModel, market, pricingModel, grantStack, founder, locale } Output:

• assumptions[] (label, value)
• revenueProjection[] (year, customers, arr, revenue)
• costStructure[] (category, y1, y2, y3)
• fundingStack[] (source, amount, stage)
• keyMetrics[] (metric, value) // CAC, LTV, gross margin, runway, breakeven
• milestones[] (month, milestone, arrTarget)
• summary

UI: projection table + funding-stack chips + key-metric cards.

---

Shared mechanics

• Every feature has a mock fallback (no API key) clearly labelled DEMO.
• locale added to FounderProfile-adjacent payloads → system prompt tells the model to answer in EN or Darija.
• New "AI Studio" section hosts all three tabs, pre-filled from src/data/workspace.ts.

Build order (this session)

1. Types → 2. Server (schemas + prompts + mocks + dispatch) → 3. Client → 4. UI components → 5. AI Studio section + nav → 6. build.