⚠ Synthetic data · not a real client Every document, name, SSN, EIN, employer and dollar figure below is invented for this demo. No client file, real or otherwise, is involved.
Document Engine
DEMO FIRM · SYNTHETIC SET ⚠ SYNTHETIC DATA ENGINE: — COGS — of —
Synthetic data · not a real client

Document pipeline

Every document that hits a firm — email, client portal, front-desk scanner — makes the same trip. The engine does the repetitive 80%; your people verify 100%. This runs entirely in your browser on ten invented tax documents. Nothing is sent anywhere.

01
Hit Receive next batch and watch documents get read, classified and extracted.
02
Go to the Human gate. Nothing files until a person approves it.
03
One document is misread on purpose. Catch it before you approve.
04
Then search what you filed, and read the cost ledger for every call.
In pipeline
0
Awaiting human
0
Filed & searchable
0
Engine cost so far
$0.00

📥 Inbox

🏷 Classifying

⚡ Extracting

✋ Human gate

🗄 Filed

Synthetic data · not a real client

Human approval gate

Nothing gets filed until a person says so. The engine shows its work and flags what it isn't sure about — low-confidence fields light up red.

⚑ The moment this demo exists for
In the second batch, a handwritten 1099-NEC comes in. The engine reads the handwritten box-1 figure as 8,750.00 at 62% confidence — the document actually says 8,150.00. It flags itself, hands you the document alongside its read, and waits. You fix the field and approve. That correction is what the audit trail keeps. It drafts. You decide.
Select a document from the queue.
Synthetic data · not a real client

Instant retrieval

Once filed, every document answers by name, type, employer, or amount — no drilling through folders. Retrieval reads the extracted index, so it costs nothing per search.

Synthetic data · simulated token counts

Usage ledger SIMULATED COUNTS · REAL UNIT PRICES

Every engine call is metered: model, tokens, cache hits, cost. Every TLW deployment includes this — the day-30 pilot report comes straight off this page. In this demo the token counts are simulated (no model is called); the per-million unit prices are the real published Claude API prices.

Engine calls
0
Docs processed
0
Cost per document
Month-to-date COGS
$0.00
Budget position — alarms at 50 / 75 / 90% 0%
50
75
90

Tax-season projector

Adjustable assumptions — you set these numbers, not us. The engine cost is computed from this demo's measured cost per document; the hours and labor value are arithmetic on the assumptions you type in. Nothing here is a claim about your firm.
Peak-month engine cost
Peak-month staff hours returned
Labor value / peak month
TimeTaskModelDocument Tok inTok outCache read CostSource