ITEMIZE / DOCUMENT AIchapter 1 · problem

Enterprise banks, payment processors, and healthcare customers needed financial documents classified at production accuracy.

0.00%
PRODUCTION ACCURACY
0M+
DOCUMENTS / YEAR
0K+
AUTOMATED MONTHLY
ITEMIZE / DOCUMENT AIchapter 2 · constraints

The honest part.

One product manager

No dedicated engineers, no data-science team on call. Every prompt, eval, and architecture decision was mine to make and mine to fix: shipped by directing AI, not by handing off.

One week

The window between commitment and the first production traffic. The schedule shaped the architecture: buy nothing, build the smallest thing that can be trusted.

Confidentiality

Itemize's customer names stay confidential. The numbers, the architecture, and the process are real, and verifiable in conversation.

ITEMIZE / DOCUMENT AIchapter 3 · prd snapshots

The paper trail, redacted.

Real documents drove this: PRD, eval plan, rollout memo. Client specifics stay under NDA; the process doesn't.

PRD v1 · problem & hard gate
       REDACTED · NDA

How success was scored: final-output grading on single-shot extraction, one answer checked against hand-labeled ground truth. Not agent-trajectory scoring. Calling these agent evals would be overselling them.

Eval plan · golden set design
      REDACTED · NDA

100 documents per type across roughly 100 document types (invoices, receipts, statements, tax forms, healthcare forms, non-invoice documents), roughly 10,000 in the golden set. I hand-labeled all the ground truth; a business analyst double-checked every set. Built or refreshed at every customer onboarding, whenever a new layout showed up in production, and monthly regardless.

Rollout memo · exceptions path
     REDACTED · NDA

The bar every version had to clear before shipping: 99% per field on the golden set, plus zero regressions against the previous version. Fail either and the release was blocked. I held the go/no-go on every rollout.

ITEMIZE / DOCUMENT AIchapter 4 · architecture

Three models vote. One gate decides.

architecture · llm ensemble → deterministic hard gatedrag →
LLM ENSEMBLEmodel · amodel · bmodel · cconsensus voteagreement + confidencedeterministic hard gateauto ✓human reviewexceptions only

models drawn from: anthropic · openai · gemini · aws in-house · ocr: amazon

ITEMIZE / DOCUMENT AIchapter 5 · ai workflow

Accuracy was climbed, not found.

prompt iteration · v3 → v5.4, accuracy climbingdrag →91.2%
92%94%96%98%91.2%v3single model, zero-shot94.8%v4few-shot + taxonomy rewrite97.3%v5ensemble vote98.51%v5.4+ deterministic hard gate
ITEMIZE / DOCUMENT AIchapter 6 · iterations

Before / after.

drag the handle · arrow keys work too

BEFOREmanual queue
  • Every document touched by a human
  • Hours of turnaround
  • Accuracy capped by fatigue
AFTERautomated pipeline
  • Humans see only flagged exceptions
  • Seconds of turnaround
  • 98.51% in production, measured not vibes
ITEMIZE / DOCUMENT AIchapter 7 · production

The day I stopped the line.

A healthcare customer needed 26 line-item fields and 7 header fields extracted from claim forms, delivered to their production SFTP as a zip: pipe-delimited CSVs, plus every claim split into single-page PDFs tagged with GUIDs.

A personal habit, not a ticket: about two hours a day auditing random documents in production. One of those audits caught 400 to 500 of every 1,000 documents landing in the low-confidence queue at under 60% accuracy, while the rest ran around 95% per line-item field on the same claims. I stopped the production line myself, before the customer saw it. Then I sized the problem: a hand audit of roughly 400 failing documents, three buckets.

~80% · Scanned below 300 DPI

Asked the customer to rescan, best effort. Then changed extraction to a two-way check: Amazon OCR output compared against the LLM extraction, with the LLM making the final call per field. That lifted even the best-effort low-quality documents to the 99% per-field bar.

~10% · Live PDFs losing form data at ingestion

The model was seeing a blank claim form. The developers added a PDF flattening script at the ingestion stage, so form data is never dropped before extraction.

~10% · Blurry images, overlapping digits

Routed to human review, by design. Not every document should be automated.

<60% affected documents, at the catch84% mid-climb99% bar back at the bar, about two weeks

Today under 5% of this customer's claims route to human review, decided by one number: a per-document confidence score below 90%.

ITEMIZE / DOCUMENT AIchapter 8 · live metrics
0.00%
PRODUCTION ACCURACY
0M+
DOCUMENTS / YEAR
0K+
AUTOMATED MONTHLY
ITEMIZE / DOCUMENT AIchapter 9 · what i'd do next

What I'd do next.

The hard gate is deliberately conservative: it routes anything ambiguous to humans, which costs automation rate. The next lever is a confidence-calibrated gate trained on six months of production disagreements, which should recover most of the flagged volume without giving back accuracy. After that: the same pattern, applied to the next document family.