Rules read symptoms
Thresholds and velocity checks fire on what a single account did. By then the money has already moved through nine others.
Financial Genome Infrastructure
Fraud does not live inside a transaction. It lives in the relationships between them. Phinome holds the financial system as one living graph — 1.3 terabytes of it, in memory, evolving with every event — and reads the hidden traits that no rules engine can see.
01 — The thesis
Thresholds and velocity checks fire on what a single account did. By then the money has already moved through nine others.
Classical scoring needs a past. A three-day-old synthetic identity has none — which is precisely why it was created.
We sequence the connective tissue: shared markers, funding paths, timing signatures. Structure betrays intent long before behaviour does.
02 — Core infrastructure
The double helix holds the code of life. Helix holds the code of money.
A single in-memory graph of every entity in your system — customers, devices, instruments, counterparties, sessions — held as a dense integer map rather than a document store. No adjacency lookups over the network. No nightly rebuild. The strand is always current, and always resident.
We don’t store transactions. We sequence the relationships between them, in real time.
Simulated strand — pulses show funds traversing a mule chain toward a sink node.
03 — The suite
Every product reads the same live graph. Nothing is exported, re-indexed or reconciled — each API simply asks the genome a different question.
Synthetic identity defence
In biology a chimera is one organism carrying the cells of two individuals. A synthetic identity — a real SSN spliced to a fabricated name — is exactly that. Chimera ID watches for a single genetic marker expressing itself across multiple, supposedly disconnected identities.
POST /v1/chimera/evaluate
{
"purity_score": 0.19,
"verdict": "spliced_identity",
"shared_markers": ["ssn_hash", "device_id"],
"collides_with": 4
}
Mule & laundering detection
A vector doesn’t cause the disease — it carries it. Money mules are the vectors of the financial system, ferrying dirty funds toward a sink. Vector Trace follows velocity and path, flagging infection chains where value moves too fast through nodes with no other economic purpose.
GET /v1/vector/trace/tx_8fd21
{
"contagion_level": 87,
"chain_depth": 6,
"dwell_time_avg_s": 42,
"terminal_sink": "acct_00b41c"
}
Inductive risk prediction
The genotype is the code; the phenotype is how it expresses. A fraudster can hold a spotless genotype — valid documents, clean history — while behaving exactly like a known ring. Our inductive graph neural network reads a new account’s environment to predict expression before any history exists.
POST /v1/phenotype/express
{
"risk_expression": 0.93,
"account_age_days": 2,
"expressed_trait": "bust_out_ring",
"neighbourhood_hops": 3
}
Active defence layer
Detection without intervention is archaeology. CRISPR Block is the response layer: instead of flagging a transaction for tomorrow’s queue, it returns a cut signal to your ledger before the money moves — excising the bad strand from the block.
POST /v1/crispr/authorize
{
"action": "cut",
"latency_ms": 6.8,
"basis": ["chimera", "vector"],
"reversible_until": "T+0s"
}
Gene editing for your transaction ledger — excise fraud before it replicates.
04 — The platform
One pass over the strand. No queue, no warehouse, no overnight window.
Events stream in from your core, KYC vendors and device SDK. Each becomes a node or an edge on the strand within microseconds — never a batch load.
event → strand.append()
Helix resolves the entity against every marker it already holds, collapsing aliases and surfacing collisions the moment they appear.
helix.sequence(entity)
Chimera, Vector Trace and Phenotype read the neighbourhood in parallel and return purity, contagion and expression in a single decision envelope.
suite.express(node) → scores
CRISPR Block returns cut, hold or pass inside your authorization call — with the full evidence path attached for your analysts and your regulator.
crispr.authorize(tx) → "cut"
of synthetic identities in benchmark portfolios surfaced before first credit line.
median inline decision, measured at the authorization boundary.
fewer false positives than velocity-rule baselines at equal catch rate.
of traversal depth available to every scoring call, in real time.
Figures from design-partner simulations on anonymised portfolios. Your mileage will depend on your strand.
05 — Early access
We onboard a small number of design partners each quarter. Tell us what you’re seeing and we’ll come back with a read on where a live graph would change the outcome.
Typical reply time — under two business days.
Thanks — your details are with us. Expect a reply within two business days.