The prototype
Working on Google Cloud today.
Across Asia Pacific, banks are told to freeze scam money faster. FairFreeze runs everything after the freeze: it holds only the disputed money, clears innocent customers in hours, stops mules faking innocence, and drafts every letter in the customer's language.
One officer, four cases, three countries. Everything you see is the working prototype on Google Cloud.
The demo bank is loaded with four synthetic cases. Ravi's is left untouched so you can run every step.
Open the officer console. Press step 1, then "Use the sample evidence", then "Run the Case Agent" and watch it work.
Click Vikram. His documents pass every check, yet his chat matches 14 cases at 3 banks. Notice there is no customer letter.
Open Ravi's link in Hindi, or Nok's in Thai, on your phone.
Open Regulator view at the top of the console and compare the Proportionality Index before and after.
Fraud engines find mules and trigger freezes. What happens next is manual, slow and often wrong: whole accounts locked over small amounts, sellers who did nothing wrong stuck for weeks, victims waiting for money that sits in someone else's account.
of savers stuck behind frozen accounts; courts now tell banks to hold only the disputed amount.
accounts frozen in 13 months; innocent customers must be released within hours of verification.
restriction orders on likely scam victims, renewable; daily living money must stay available.
AFCA can now hear complaints against receiving banks about scam money.
Gemini reads and reasons. Deterministic code does every number. A bank officer decides.
Police notice, court order, AOC or restriction order, in any language and format.
Account, holder, transaction, amount and legal basis against the bank's own record and the country rule pack.
The lowest-intermediate-balance rule shows what is provably the victim's money: hold that, release the rest, find the next hops.
A secure link in Hindi, Thai, Mandarin or English shows what is held and what is free, and collects the sale chat and receipts.
Cross-checks with the bank record, behaviour signals, and reuse detection across other banks' cases.
Gemini 4 Argon chooses its own tool calls, writes the assessment, the police memo and the customer letter with rule citations.
Decision, audit trail in Firestore, and the regulator's Proportionality Index.
A number every regulator can rank banks by. In Ravi's case:
Whole-account freeze: INR 84,000 held for an INR 3,600 dispute.
With FairFreeze: hold INR 3,600, release INR 80,400 the same day.
INR 3,600 from a buyer whose money came from a scam. Police asked for a whole-account freeze.
One THB 1,250 payment disputed. Thai notice read straight from the document.
His chat and dates all match. But the same script appears 14 times at 3 banks, the money left for Dubai and Hong Kong within minutes, and 92% of logins come from Cambodia.
72, under a restriction order while an "investment mentor" asks for SGD 60,000.
Cheap, fast models where the task is reading. The strongest reasoning model where judgement is needed. No model at all where the answer must be exact.
We see frozen-account complaints from the inside. FairFreeze is the tool we wished our banks had.
Product, business analysis and cloud
Business analysis
Development
Quality assurance