Phase 1: Setup

Use ultrathink to thoroughly analyze performance bottlenecks and optimization opportunities.

Context: Next.js 14 (App Router) frontend + FastAPI backend. Running locally via pnpm dev (port 3000) and uvicorn (port 8000). Auth: Cookie-based session (login as test user before benchmarks).

Goal: Establish reproducible performance benchmarks, identify top 3 optimization opportunities, implement them, and measure impact.

Success criteria: Achieve ≥20% improvement in LCP or ≥30% reduction in JS bundle size, whichever comes first. Stop after 3 optimization rounds regardless.

Phase 1: Setup

1. Install: playwright, lighthouse, @next/bundle-analyzer

2. Create benchmarks/ with this structure:

   benchmarks/
     scripts/
       lighthouse.ts    # Core Web Vitals via Lighthouse
       api-timing.ts    # FastAPI endpoint latency
       bundle-size.ts   # Analyze next build output
     results/
       baseline.json
     config.ts          # Routes, endpoints, run count
   

3. Define targets in config.ts:

   • Routes: /, /dashboard, /candidates/[id]
   • API endpoints: GET /cvs, POST /cvs/{id}/match, GET /jobs
   • Runs per measurement: 5 (discard first as warmup)

Phase 2: Baseline

Measure and record (use median of 5 runs):

Output: baseline.json with structure:

{
  "timestamp": "ISO8601",
  "routes": { "/dashboard": { "lcp_ms": 1850, "ttfb_ms": 320 } },
  "api": { "GET /cvs": { "p50_ms": 45, "p95_ms": 120 } },
  "bundle": { "total_kb": 412, "largest_chunks": [...] }
}

Phase 3: Analyze

Output analysis.md with:

1. Top 3 bottlenecks ranked by impact (quantified)
2. Root cause for each (e.g., "lodash bundled entirely for one function")
3. Proposed fix with estimated effort (S/M/L)

Phase 4: Optimize Loop

For each fix (max 3 iterations):

1. Create branch: perf/optimize-{description}
2. Implement fix
3. Run pnpm build && pnpm test — abort if tests fail
4. Re-run benchmark suite
5. Append to results/optimization-{n}.json
6. Commit with message: perf: {description} (LCP -X%, bundle -Y%)

Stop when: target met OR 3 iterations complete OR next-best opportunity offers <5% projected improvement.