A national health services organisation was reconciling member records by hand. We built a LangGraph multi-agent platform on Databricks that resolves every incoming member event into one governed, auditable golden identity — and escalates to a human steward whenever confidence falls below the threshold.
Build Your Identity Resolution PlatformFor a healthcare payer, few problems are as deceptively simple as knowing with certainty that the member in front of you is the same member represented across a dozen other systems. When identity is uncertain, everything built on top of it inherits that uncertainty: eligibility checks, claims adjudication, care coordination and member outreach all risk acting on the wrong record, or on a fragment of the right one. The platform resolves incoming member events into a Member Identity Graph — a continuously updated representation of true member identity that eligibility engines, claims processors and next-best-action systems can depend on as a shared source of truth.
A national health services organisation ingesting member data from enrollment platforms, claims feeds, provider networks and third-party administrators. The source does not name the client.
Each source captures identity slightly differently — a name spelled one way here, a date of birth transposed there, an identifier present in one feed and absent in another.
Extraction, matching, scoring, escalation and resolution are coordinated as one workflow with the state, memory and control flow needed to handle exceptions gracefully rather than forcing every case through an identical pipeline.
Because identity resolution is exposed as a dependable governed service, the client's teams are extending the same platform to an Intent Manager Agent — the first downstream capability built on the infrastructure rather than beside it.
Every agent operates against a BaseTool contract, every output is validated against policy before it is acted upon, and every decision lands in a trail from which the reasoning can be reconstructed on demand.
Scope an agentic AI platform where governance, auditability and human oversight are architectural requirements rather than compliance afterthoughts.
Manual reconciliation is slow, expensive and fundamentally unable to scale to the event volumes a national health services business generates.
Agentic AI deployed at the core of a regulated operation, not bolted onto the edge of it.
Identity resolution is a sequence of judgment calls, each needing different tools, thresholds and escalation paths. A multi-agent workflow handles that; a linear pipeline forces every case through the same brittle path.
The human-in-the-loop path is built on durable interrupts with preserved state, not a queue bolted on afterwards. The system extends human judgment in ambiguous, high-stakes cases rather than replacing it.
Immutable audit trails, policy-as-configuration governance and restart-safe checkpointing were first-class requirements, which is what lets the organisation reconstruct any decision for an auditor on demand.
The identity graph was built to be extended. The Intent Manager Agent now being built on it is the test of that: a capability that compounds in value rather than depreciating into another isolated tool.
An ambient clinical documentation agent that captures the encounter and drafts the note, with the clinician as author and signer.
Shift scheduling, NFC and geofenced clock in, CMS PBJ compliance, in house payroll and embedded banking across five client apps.
