A forward deployed pod of 3–5 tkxel engineers builds AI agents for prior authorization, clinical documentation and revenue cycle work inside your environment, under a BAA, with minimum-necessary PHI access, clinician review and audit logs, then hands it over.
Book discovery callAgents check payer rules for the ordered service, pull supporting notes, labs and imaging from the EHR, assemble the request for the payer portal or FHIR Prior Authorization API, and track status. Clinical judgement calls and peer-to-peer reviews stay with your staff.
Agents draft visit summaries, referral letters and discharge instructions from encounter data, and suggest ICD-10 and CPT codes with the note text that supports each one. Clinicians sign every note; coders accept or reject suggestions, and their reasons feed back into testing.
Agents read 835 remittance advice and denial codes, group denials by root cause, pull the chart evidence that answers the payer’s reason and draft the appeal letter. Your billing team reviews and submits, and denial trends go back to front-end registration.
Before scheduled visits, agents run 270/271 eligibility checks, read plan documents for coverage limits and authorization flags, and post results to the registration record, so front-desk staff see problems days ahead rather than at check-in.
For health plans, agents classify incoming authorization requests, check completeness against medical policy, extract clinical criteria from attached records and prepare a case summary for the nurse reviewer. Agents never issue denials; adverse determinations stay with licensed clinicians.
For digital health companies, the pod builds the AI features and FHIR or HL7 v2 integrations your roadmap needs, with tenant-level PHI isolation, de-identified data for model testing and logging designed to stand up to a HITRUST or SOC 2 assessment.
We’ll outline data flows, BAA scope and reviewers for one workflow in 30 minutes.
Book a scoping callSecurity reviews, BAAs and EHR access requests often outlast a pilot budget before any engineering starts.
Calling a model is easy. Getting reliable data out of Epic, Oracle Health or a payer core system, and writing results back, is where projects slow down.
A draft note or authorization summary without cited source text gets skipped, so adoption fades after the demo.
Medical policies, CMS timelines and portal formats shift, and a system nobody on staff can update degrades quickly.
We agree the workflow, the minimum PHI each step needs and the human reviewer for every decision, and execute a BAA covering tkxel and every model provider in the data flow. Your privacy and security teams review the data flow before access is granted.
Deliverables: BAA, PHI data flow diagram, decision boundary, access request
The pod builds inside your cloud tenancy or integration layer using BAA-covered model endpoints, with encryption, role-based access and per-request audit logging. Testing starts on de-identified or synthetic records before any live PHI is used.
Deliverables: working agent, FHIR and HL7 integrations, evaluation set, audit log design
The agent runs on live cases with every output reviewed by your clinicians, coders or authorization staff, and their corrections become test cases. We report accuracy by case type so you can decide where review can lighten.
Deliverables: pilot results, error analysis, monitoring thresholds, security risk analysis update
Your engineering and revenue cycle teams take ownership with runbooks, payer-rule update procedures and on-call shadowing. Code, prompts and tests stay in your repositories, and our access is revoked on exit.
Deliverables: runbooks, source repositories, rule update guide, training sessions
We build in your repositories and environments from day one, so there is no separate rebuild between proof of concept and release.
Each pod is led by an engineer who has shipped AI systems before, drawn from tkxel’s team of around 150 AI professionals. The same people stay through delivery.
Autonomy matrices, audit trails, and human review are designed with your risk and security teams before release, backed by ISO 27001 controls.
Every engagement starts with agreed business metrics and evaluation sets. We report against them throughout, and the pod exits when your team can run the system.
AI in healthcare operations mostly targets the administrative work around care rather than the clinical decision itself: assembling prior authorization requests, drafting documentation, suggesting codes, verifying eligibility and working denials. Agents read from the EHR, practice management and payer systems, prepare the work and hand it to a person to approve. Clinical decisions, adverse determinations and note sign-off stay with licensed staff, and that is how we design every system.
Partly. Agents can check whether a service needs authorization, assemble the supporting documentation, and submit and track requests, but medical necessity disputes and peer-to-peer calls still need clinicians. According to the CMS fact sheet for the Interoperability and Prior Authorization Final Rule (CMS-0057-F), impacted payers must offer a FHIR Prior Authorization API, with API compliance dates generally beginning January 1, 2027. Building on FHIR now avoids rework, though portal submissions will remain common with many commercial payers for some time.
It can be, if the model provider signs a business associate agreement, PHI stays on covered infrastructure, access follows the minimum necessary standard and every use is logged. HIPAA does not certify tools; compliance depends on how the system is configured and run. We use only BAA-covered model endpoints, keep PHI in your tenancy and document the data flow for your privacy officer and security risk analysis. Your counsel should still review subprocessor terms.
Yes. Because our engineers work inside systems that hold PHI, we sign a BAA before any access is granted and confirm that every model provider or subprocessor in the data flow is covered by its own agreement. Engineers complete HIPAA training, work on your managed devices or virtual desktops where you require it, and lose access the day the engagement ends. tkxel holds ISO 27001 certification, which your vendor review can reference.
They struggle where records are scanned, handwritten or split across systems the agent cannot reach, and where payer rules are unpublished or inconsistent. Accuracy in rare specialties is lower because there are fewer cases to test against. We will tell you during scoping if a workflow’s data quality makes automation unreliable; sometimes the honest recommendation is to fix the intake form or the interface first and add the agent later.
For one workflow, expect roughly 12–16 weeks from kickoff to handover, with production use starting during a supervised pilot. The longest delays are rarely the build: BAA execution, security review and EHR interface access commonly take several weeks, so we start them on day one. Payers with complex utilization management policies, or providers integrating several EHR instances, should plan for the upper end of that range or beyond.
Governance, Agents
1 October 2026 · 8 min read
Read article
Governance
24 September 2026 · 8 min read
Read article
Agents, Governance
17 September 2026 · 8 min read
Read article