← Back to Registry Hub/Site Search/L3 · capstone_project
⚡Flashcard Drill➔
L3S2 · Data & RetrievalS3 · Model & AgentS4 · UX & HITLS5 · Eval & LLMOpsS6 · Security & FinOpsVerified: 2026-11-15

HIPAA-Compliant Clinical Prior Authorization Evidence Agent (Capstone Project)

Reproducible healthcare AI capstone using Synthea synthetic FHIR records and public payer policies: compiles boolean medical-necessity criteria graphs, retrieves mixed FHIR/PDF evidence under PHI scrubbing rails, and gates releases via a 50-case clinical eval suite.

Best For

Candidates targeting Healthcare AI, Applied AI, or AI Engineer roles needing to prove mastery of regulated, audit-grade agent systems over synthetic FHIR data.

Human-AI Boundary

Renders Met / Evidence Missing / Clinician Attestation badges with one-click export of an annotated audit packet for physician signature.

Anti-Patterns to Avoid

  • Training or testing capstone demos on real unmasked patient records from public scrapes
  • Building a plain chat wrapper without deterministic checklist verification of ICD-10 / CPT and date/threshold rules

Latency/Cost Profile & Eval Metrics

Async case evaluation ~6-12s E2E on local/cloud OpenAI-compatible runtime; interactive HITL inspection < 150ms · Policy graph pre-compilation (once per policy) + per-case FHIR/PDF checklist verification

  • Medical Necessity Criterion Sensitivity & Specificity: >= 0.94 sensitivity / zero false-Met endorsements on missing prerequisite labs (Evaluated across 50 synthetic cases (25 eligible, 15 incomplete, 10 edge conflicts))
  • PHI Leak Redaction Precision: 100% synthetic MRN/DOB/SSN redaction before trace export (Verified by automated Presidio/regex telemetry interceptor tests)

💰 Business ROI Model & Evidence Level

portfolio_blueprint

🏗️ End-to-End Architecture Pipeline

Step 1: Synthea FHIR Loader & Docling Referral Parser

Load structured FHIR resources and parse synthetic fax PDFs into coordinate-indexed chunks.

Step 2: Criteria DAG Verifier & Zero-PHI Telemetry Gate

Run LangGraph per-criterion verification with mandatory Presidio PHI scrubbing prior to trace logging.

Step 3: 50-Case Eval CI Gate & Co-Audit Split UI

Benchmark accuracy delta vs naive prompting and expose clickable page anchors for clinician attestation.

📦 Project Deliverables & Portfolio Artifacts

  • Policy-to-Criteria Graph Compiler — Compiles public payer coverage PDFs into versioned AND/OR checklist JSON trees
  • Zero-PHI Telemetry & 50-Case Eval Harness — Includes PHI redaction unit tests, missing-evidence adversarial cases, and Ragas regression CI

🎤 Interview Defense & System Design Deep-Dive Q&A

  1. When comparing a single 100k-token prompt containing full patient history + policy vs your Criteria DAG architecture, what were the exact false-positive failure modes?
⚡
Test muscle memory on this topic via 3-Layer Flashcards
Active Recall: Mentally formulate Prompt ➔ Math Derivation ➔ Benchmark Implementation in 3 minutes.
🎯 Drill This Card Now➔

Bidirectional Taxonomy Links

[HubB] Role Lens · Applied AI & Agent Systems Engineer[HubB] Role Lens · Machine Learning & Inference Systems Engineer[HubB] Role Lens · AI Decision Science, Experimentation & Eval Data Scientist[HubB] Role Lens · AI Product Manager & Applied Strategy Lead[HubB] Role Lens · Backend / Full-Stack SWE Pivoting into AI Systems[HubB] Role Lens · Forward Deployed Engineer (Enterprise AI Delivery & Solutions)[L0 Card] LangGraph State Machine[L0 Card] Citation Grounding & Faithfulness[L0 Card] NeMo Guardrails Programmable Rails[L0 Card] RAG Triad Evaluation (Ragas)[L0] Docling Layout & Complex Table PDF Parser[L0] RAG Triad Faithfulness Metrics & Multi-Hop Hallucination Failure Primitives[L0] LLM Token FinOps, Prefix Caching & Tenant-Isolated Semantic Cache Gateway[L1] Complex Long-PDF Layout Parsing & Hybrid Reranked Retrieval Pattern[L1] Stage 3 · Bounded State-Machine & Plan-and-Execute Tool Orchestration Pattern[L1] Stage 4 · Inline Grounded Citation Streaming & HITL Escalation UX Pattern[L1] Stage 5 · Golden-Dataset CI/CD Regression Gate & Trajectory LLMOps Pattern[L1] Stage 6 · Multi-Tenant Vector ACL Isolation, Prompt Firewall & FinOps Breaker Pattern[L2] Clinical Prior Authorization Evidence Synthesis Copilot Playbook

Internal Tools & Primary Sources