Return to presentationUC2 - Intelligent Referral Matching and RoutingSynthetic data

Deterministic POC demonstration using synthetic data

UC2 - Intelligent Referral Matching and Routing

Read, search, filter, rank, explain, and route. Network participation and accepting-new-patient filters run before ranking.

Choose a synthetic scenario

Request payload

POST /referrals/match

{
  "referral_id": "REF-002",
  "member_id": "M002",
  "referral_text": "Chronic knee pain with failed conservative therapy. Refer to knee specialist / orthopedic surgeon.",
  "zip_code": "22180",
  "product": "PPO"
}

Workflow stages

  1. 01Read

    Referral intake and structured extraction.

  2. 02Search

    Provider tool returns authoritative candidates.

  3. 03Filter

    Deterministic network and new-patient filters run before ranking.

  4. 04Rank

    Only valid candidates are ranked.

  5. 05Explain

    Evidence, rules, score, and confidence.

  6. 06Route

    Route the referral or send it to human review.

Control point. Deterministic network and accepting-new-patient filters run before ranking. Only valid candidates reach the ranking step.

ROUTEDReferral REF-002Confidence 0.93Human review false

Interpreted need

Specialties: Orthopedic Surgery
Focus: Knee
Ambiguous: false
Confidence: 0.93

Rules applied

  • Product network participation required
  • Accepting new patients required
  • Hard filters precede ranking

Excluded before ranking

ProviderSpecialtyNetwork productsExclusion reason
P101, Metro Knee InstituteOrthopedic SurgeryHMONot in member product PPO

Ranked candidates

ProviderSubspecialtiesDistance (mi)QualityMatch score
P100, Capital Ortho Sports MedicineSports Medicine, Knee4.24.60.951
P102, Northern Virginia OrthopedicsGeneral Orthopedics7.14.30.723

Explanation

Capital Ortho Sports Medicine is the highest-ranked valid candidate after network and new-patient filters. The ranking uses specialty/subspecialty match, quality, and distance.

Audit events

  1. Referral received
  2. Referral interpretation invoked
  3. Provider tool returned 3 candidates
  4. Deterministic network/new-patient filters applied before ranking

Saved response - demo_outputs.json

{
  "referral_id": "REF-002",
  "status": "ROUTED",
  "interpreted_need": {
    "specialties": [
      "Orthopedic Surgery"
    ],
    "focus": [
      "Knee"
    ],
    "ambiguous": false,
    "confidence": 0.93
  },
  "rules_applied": [
    "Product network participation required",
    "Accepting new patients required",
    "Hard filters precede ranking"
  ],
  "excluded_candidates": [
    {
      "provider": {
        "provider_id": "P101",
        "name": "Metro Knee Institute",
        "specialty": "Orthopedic Surgery",
        "subspecialties": [
          "Knee",
          "Joint"
        ],
        "zip_code": "22031",
        "network_products": [
          "HMO"
        ],
        "accepting_new_patients": true,
        "distance_miles": 6,
        "quality_score": 4.9
      },
      "reasons": [
        "Not in member product PPO"
      ]
    }
  ],
  "ranked_candidates": [
    {
      "provider_id": "P100",
      "name": "Capital Ortho Sports Medicine",
      "specialty": "Orthopedic Surgery",
      "subspecialties": [
        "Sports Medicine",
        "Knee"
      ],
      "zip_code": "22180",
      "network_products": [
        "PPO",
        "HMO"
      ],
      "accepting_new_patients": true,
      "distance_miles": 4.2,
      "quality_score": 4.6,
      "match_score": 0.951
    },
    {
      "provider_id": "P102",
      "name": "Northern Virginia Orthopedics",
      "specialty": "Orthopedic Surgery",
      "subspecialties": [
        "General Orthopedics"
      ],
      "zip_code": "22182",
      "network_products": [
        "PPO"
      ],
      "accepting_new_patients": true,
      "distance_miles": 7.1,
      "quality_score": 4.3,
      "match_score": 0.723
    }
  ],
  "explanation": "Capital Ortho Sports Medicine is the highest-ranked valid candidate after network and new-patient filters. The ranking uses specialty/subspecialty match, quality, and distance.",
  "confidence": 0.93,
  "human_review_required": false,
  "audit": [
    "Referral received",
    "Referral interpretation invoked",
    "Provider tool returned 3 candidates",
    "Deterministic network/new-patient filters applied before ranking"
  ]
}

Implementation boundary

  • The current POCs use local deterministic implementations to test workflow contracts, fixed controls, task structure, tool boundaries, and exception paths.
  • They do not call an LLM or Amazon Bedrock.
  • A production architecture can introduce an approved model gateway, such as a Bedrock-based gateway, without changing task and tool contracts.
  • This web experience replays the exact saved outputs of the supplied POC package. It is not an LLM result, a Bedrock result, or a production result.

The reasoning layer never invents a provider or changes network status.