Cash Flow Anomaly Detector

How Cash Flow Anomaly Detector works

Press Load model once: your browser downloads Gemma 4 E2B instruction-tuned (web build) (2.0 GB) from this site and keeps it in its cache. Then add your text and press Classify locally. The model runs on this device using your graphics card, so nothing you enter is sent to a server. Until the model is loaded, or if your device cannot run it, the tool falls back to a simple keyword check and labels the result as the fallback.

Gemma 4 E2B instruction-tuned (web build) is a small open language model from Google. The tool sends it the instruction below plus your text, asks for a JSON answer with fixed fields, and shows each field as a card. It runs at temperature 0, so the same input gives the same answer on the same device, though different graphics cards can word it slightly differently.

Inputs longer than 6,000 characters are cut to that length before they reach the model, and the result says so. The model only sees what you paste: it does not look anything up.

The instruction the model gets

You read a list of account transactions and point out the ones that look unusual and deserve a second look: charges from payees the person may not recognise, items that look like duplicates of a normal payment, unexpected fees, deposits or debits with no clear source, and anything described as unknown or late. Say what makes each one stand out in words, using the payee, amount and date exactly as written. Do not accuse anyone of fraud; say what to check. This is not financial, tax or legal advice, so do not tell the user what they should do with their money. Do not do any arithmetic: copy every amount, date and percentage exactly as written, and never compute totals, differences, gaps, averages or percentages.

What it returns

• Items to look at: Each unusual transaction, written as 'payee and amount as written → why it stands out'.
• Possible duplicates: Transactions that look like a repeat of another one, quoting both as written.
• Routine items: Transactions that look normal and recurring (pay, rent, utilities), as written.
• What to check: Two or three concrete things to confirm with the bank, the payee or your own records.

Worked example

The model's output for the built-in sample, recorded on a desktop browser. Your run may word things slightly differently.

Checking export: payroll +$5,240, rent -$2,150, utilities -$286, Stripe deposit +$1,180, unknown ACH debit -$940 from ACME HOLDINGS, duplicate insurance draft -$183 two days after normal payment.

Items to look at

unknown ACH debit -$940 from ACME HOLDINGS → Check the source of this debit. Stripe deposit +$1,180 → Verify the source and purpose of this deposit. duplicate insurance draft -$183 two days after normal payment → Check the timing and purpose of this draft.

Possible duplicates

Stripe deposit +$1,180 duplicate insurance draft -$183

Routine items

payroll +$5,240 rent -$2,150 utilities -$286

What to check

Confirm the source and purpose of the ACH debit -$940 from ACME HOLDINGS. Verify the source and purpose of the Stripe deposit +$1,180. Check the timing and purpose of the duplicate insurance draft -$183.

The model

The files are served from this site, versioned so a cached copy is never stale. Nothing is fetched from a third party while you use the tool.

Limitations

• Small language models make mistakes: they can miss an item, merge two, or state something more confidently than the text supports. Check the result against your input before you rely on it.
• It needs WebGPU and about 8 GB of memory, so it is offered on desktops and laptops, not on phones and tablets. Devices that cannot run it get the keyword fallback, clearly labelled.
• Results describe the input you gave, not the real-world situation behind it. Review them yourself before acting.
• This is not financial, tax or legal advice. Confirm anything important against your own records and with a qualified professional.