An end-to-end Next.js 14 + Prisma + Tailwind MVP for auditing LLM applications. It now uses a modular audit pipeline that separates target execution, evaluation, and run aggregation while persisting structured evidence and version metadata for each run.

• Next.js 14 App Router (TypeScript)
• Tailwind CSS + shadcn-inspired primitives
• Prisma + PostgreSQL
• Recharts for charts
• zod for validation
• CSV / JSON exports (papaparse-style)

1. Install deps
   npm install

2. Configure env
   cp .env.example .env

3. Prisma setup & seed
   npm run db:bootstrap

4. Run dev server
   npm run dev
   Open http://localhost:3000

Seed creates a demo project (demo-project) and a completed audit run (demo-run). Nav link “Demo Run” opens the dashboard instantly.

• Landing page → “Create Audit Project”
• Configure target per mode (Prompt-only, Endpoint, RAG)
• Select audit categories → Run audit
• View dashboard (overall score, category bars, verdict donut, table)
• Drill into test details
• Export JSON/CSV from dashboard

• Server actions in lib/actions.ts handle create/update/run/export.
• Audit execution, scoring, and aggregation are separated under lib/audit/*.
• lib/evaluator.ts still provides the heuristic scorer, but it now sits behind a pipeline seam instead of being embedded inside the run loop.
• Charts rendered client-side (components/charts.tsx).
• Prisma schema in prisma/schema.prisma; seed cases in prisma/seed.js.

Add provider-specific executors in lib/audit/executor.ts for OpenAI, Anthropic, and production endpoint adapters. The current implementation keeps a simulated fallback plus a generic HTTP executor seam.

Hook into pdf-lib in an API route and render a summary; CSV/JSON already implemented.

• npm run dev – start app
• npm run build – production build
• npm run prisma:push – sync schema
• npm run prisma:seed – seed demo data
• npm run db:bootstrap – generate client, push schema, and seed demo data

Internal demo use. Adjust as needed.