Where this stands
Case study in progress. Target completion: 2026-12-15. The full writeup lands once the research, build, and eval phases are done.
The customer scenario
CPA firm or enterprise tax/finance team automating high-volume structured document generation needs reliable extraction, validation, and human-in-the-loop fallback.
My product canvas
This case study is grounded in TaxEXP — an automated K1 and Form 1065 tax document generation pipeline. CPA variant adapts the same engine for accounting practice workflows. Canonical writeup: kaydenlabs.com/work.
Architectural patterns to be demonstrated
- High-volume structured document pipeline — K1, Form 1065, with field-level extraction and validation
- Confidence-scored output with HITL fallback at threshold — model owns the high-confidence rows; humans own the low-confidence
- Cost modeling across Haiku / Sonnet / Opus — production routing by document complexity and field criticality
What you'll find here when it ships
- Architecture diagram
- Eval set with results
- Cost and latency analysis
- Failure modes documented
- "What I'd do differently" retro