Example: Resume Parser at Scale (1M documents, budget optimization)

[teilomillet]

Goal

Demonstrate processing massive dataset efficiently with fixed budget.

Use Case

Parse 1 million resumes with $500 total budget (~$0.0005 per resume).

What It Demonstrates

• Streaming architecture - process without loading all in memory
• Budget distribution - smart allocation across items
• p95 cost optimization - target p95 instead of worst-case
• Partial result recovery - save what works, retry what fails
• Dead letter queue - handle budget_exceeded gracefully
• Progress tracking - live dashboard of cost/throughput

Deliverables

• examples/resume_parser/
  • parser.py - streaming processor
  • budget_strategy.py - allocation logic
  • recovery.py - partial result handler
  • dashboard.py - real-time progress UI (optional)
  • README.md - architecture, cost analysis
  • cost_optimization.md - strategies and trade-offs
  • Benchmark results (mock data)

Cost Optimization Strategies

1. Dynamic allocation - give more budget to complex resumes
2. Early stopping - detect when resume is simple, use less budget
3. Batch efficiency - amortize fixed costs
4. Fallback tiers - use cheaper models for simple extractions

Success Criteria

• Simulates 1M resume processing
• Shows cost distribution (p50, p95, p99)
• Demonstrates budget exceeded recovery
• Includes optimization guide
• Benchmark: processes X resumes/sec at Y cost

Target Audience

Data teams, HR tech, high-volume processing

Estimated Effort

Large (3-4 days)