feat: Users AI Chat multi-step workflows with CDT lookup and alias management

- Add eligibility_by_id and check_and_claim intents to internal chat
- New cdt-lookup.ts: keyword search against fee schedule JSON (no LLM)
- New internal-chat-workflow.ts: deterministic orchestration — patient
  resolution, insurance siteKey derivation, CDT code mapping
- Custom CDT aliases stored per-user in DB (TwilioSettings JSON blob)
  with GET/PUT /api/ai/cdt-aliases endpoints
- Chatbot UI: new steps for eligibility-id-ready, check-and-claim-ready,
  and need-insurance-clarification with insurance picker
- Settings UI: CDT Aliases CRUD table with built-in alias reference

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/Backend/src/ai/internal-chat-graph.ts

@@ -1,39 +1,70 @@
 import { ChatGoogleGenerativeAI } from "@langchain/google-genai";
 
+// ─── Intent types ─────────────────────────────────────────────────────────────
+
 export type InternalChatIntent =
-  | "check_eligibility"
-  | "find_patient"
+  | "check_eligibility"     // by patient name → look up in DB
+  | "eligibility_by_id"     // by explicit memberId + dob (no name)
+  | "check_and_claim"       // eligibility + claim procedures
+  | "find_patient"          // look up patient record only
   | "navigate_claims"
   | "navigate_schedule"
   | "general";
 
 export interface ChatClassification {
   intent: InternalChatIntent;
-  patientName?: string;
+  // --- patient resolution (one of name OR id+dob) ---
+  patientName?: string;       // for check_eligibility / find_patient
+  memberId?: string;          // for eligibility_by_id / check_and_claim
+  dob?: string;               // for eligibility_by_id / check_and_claim (MM/DD/YYYY)
+  // --- insurance hint (only if explicitly stated in the message) ---
+  insuranceHint?: string;     // raw text, e.g. "masshealth", "BCBS", "CCA"
+  // --- procedures (raw text, NOT CDT codes — CDT lookup is done in workflow) ---
+  procedureNames?: string[];  // for check_and_claim, e.g. ["perio exam", "adult cleaning"]
   fallbackReply: string;
 }
 
-const BASE_SYSTEM_PROMPT = `You are an internal assistant for a dental office management system.
-Staff members type natural language commands and you classify what they want.
+// ─── System prompt ────────────────────────────────────────────────────────────
 
-Respond ONLY with valid JSON (no markdown, no code fences) in this exact format:
+const BASE_SYSTEM_PROMPT = `You are an internal assistant for a dental office management app.
+Staff type natural language commands. Your ONLY job is to classify the intent and extract
+structured parameters. Do NOT map procedure names to CDT codes — return them as plain text.
+
+Respond ONLY with valid JSON (no markdown fences):
 {
-  "intent": "<one of the intents below>",
-  "patientName": "<full name if patient is mentioned, otherwise omit>",
-  "fallbackReply": "<a short, helpful reply to show the user>"
+  "intent": "<intent>",
+  "patientName": "<full name if mentioned by name>",
+  "memberId": "<member/insurance ID if given explicitly>",
+  "dob": "<date of birth in MM/DD/YYYY if given>",
+  "insuranceHint": "<insurance name only if explicitly stated in the message, e.g. 'masshealth', 'BCBS MA', 'CCA'>",
+  "procedureNames": ["<raw procedure name>", ...],
+  "fallbackReply": "<1-2 sentence reply to show the user>"
 }
 
+Omit any field that is not present in the message.
+
 Intents:
-- check_eligibility: user wants to check insurance eligibility for a patient (e.g. "check MARIA", "verify insurance for John Smith", "eligibility GONZALES")
-- find_patient: user wants to look up a patient record only (e.g. "find patient John", "look up Smith", "show me GONZALES record")
-- navigate_claims: user wants to open the claims page
-- navigate_schedule: user wants to open the appointments/schedule page
-- general: anything else — answer helpfully based on dental office context
+- check_eligibility  : user wants to check insurance for a patient identified by NAME only
+                       e.g. "check Maria Jesus", "verify insurance for John Smith"
+- eligibility_by_id  : user provides a member ID and date of birth (no patient name)
+                       e.g. "check masshealth for 100xxxx, 10/10/1988"
+- check_and_claim    : user wants to check eligibility AND submit procedures as claims
+                       e.g. "check masshealth for 100xxxx, 10/10/1988 and claim perio exam and adult cleaning"
+                       e.g. "check Maria Jesus and claim D0120 D1110"
+- find_patient       : look up a patient record only, no eligibility
+                       e.g. "find patient John", "look up Smith"
+- navigate_claims    : open the claims page
+- navigate_schedule  : open the appointments/schedule page
+- general            : anything else
 
 Rules:
-- Extract the full patient name as-is from the message for check_eligibility and find_patient
-- Keep fallbackReply to 1-2 sentences max
-- For navigate intents, fallbackReply should say "Opening the [page] page..."`;
+- For check_and_claim, procedureNames should be the RAW user text
+  (e.g. "perio exam", "adult cleaning", "D0120") — do NOT translate to codes
+- insuranceHint is only set when the user explicitly names an insurance in the message
+- Keep fallbackReply to 1-2 sentences
+- For navigate intents, fallbackReply = "Opening the [page] page..."`;
+
+// ─── Classifier ───────────────────────────────────────────────────────────────
 
 export async function classifyInternalChat(
   message: string,
@@ -42,7 +73,8 @@ export async function classifyInternalChat(
 ): Promise<ChatClassification> {
   const fallback: ChatClassification = {
     intent: "general",
-    fallbackReply: "I can help you search for a patient, check eligibility, or navigate to claims or appointments.",
+    fallbackReply:
+      "I can search for a patient, check eligibility, run check & claim, or navigate to claims or appointments.",
   };
 
   if (!apiKey) return fallback;