feat: pre-fetched caller context replaces tool-based patient lookups

- CallerContextService: fetches lead profile, appointments, call history,
  activities in parallel. Caches in Redis (5 min TTL). Renders as
  human-readable KB section — no UUIDs exposed to the LLM.
- Caller resolution controller: prewarms context cache on resolve
  (fire-and-forget) so the AI stream has a cache hit.
- AI chat stream: injects caller context into system prompt KB instead
  of raw Lead ID. LLM answers patient questions from context, no tool
  calls needed for current caller data.
- Eliminates UUID hallucination: LLM never sees leadId or patientId,
  can't pass wrong ID to wrong tool parameter.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/caller/caller-context.service.ts

@@ -0,0 +1,199 @@
+import { Injectable, Logger } from '@nestjs/common';
+import { PlatformGraphqlService } from '../platform/platform-graphql.service';
+import { SessionService } from '../auth/session.service';
+
+export type CallerContext = {
+    leadId: string;
+    patientId: string;
+    name: string;
+    phone: string;
+    isNew: boolean;
+    // Lead profile
+    leadSource: string | null;
+    leadStatus: string | null;
+    interestedService: string | null;
+    aiSummary: string | null;
+    contactAttempts: number;
+    lastContacted: string | null;
+    utmCampaign: string | null;
+    // Appointments
+    appointments: Array<{
+        scheduledAt: string;
+        status: string;
+        doctorName: string;
+        department: string;
+        reasonForVisit: string | null;
+    }>;
+    // Recent call history
+    calls: Array<{
+        startedAt: string;
+        direction: string;
+        duration: number | null;
+        disposition: string | null;
+        agentName: string | null;
+    }>;
+    // Lead activities
+    activities: Array<{
+        activityType: string;
+        summary: string | null;
+        occurredAt: string;
+        outcome: string | null;
+    }>;
+};
+
+const CACHE_KEY_PREFIX = 'caller:context:';
+const CACHE_TTL = 300; // 5 minutes — covers the call duration
+
+@Injectable()
+export class CallerContextService {
+    private readonly logger = new Logger(CallerContextService.name);
+
+    constructor(
+        private readonly platform: PlatformGraphqlService,
+        private readonly session: SessionService,
+    ) {}
+
+    async getOrBuild(leadId: string, patientId: string, auth: string): Promise<CallerContext | null> {
+        if (!leadId) return null;
+
+        // Check cache first
+        const cacheKey = `${CACHE_KEY_PREFIX}${leadId}`;
+        try {
+            const cached = await this.session.getCache(cacheKey);
+            if (cached) {
+                this.logger.log(`[CALLER-CTX] Cache hit for ${leadId}`);
+                return JSON.parse(cached);
+            }
+        } catch {}
+
+        // Build fresh
+        this.logger.log(`[CALLER-CTX] Building context for lead=${leadId} patient=${patientId}`);
+        const ctx = await this.build(leadId, patientId, auth);
+        if (ctx) {
+            this.session.setCache(cacheKey, JSON.stringify(ctx), CACHE_TTL).catch(() => {});
+        }
+        return ctx;
+    }
+
+    // Fire-and-forget pre-warm — called from caller resolution
+    // so the cache is hot when the AI stream fires seconds later.
+    prewarm(leadId: string, patientId: string, auth: string): void {
+        if (!leadId) return;
+        this.getOrBuild(leadId, patientId, auth).catch(err => {
+            this.logger.warn(`[CALLER-CTX] Prewarm failed: ${err.message}`);
+        });
+    }
+
+    private async build(leadId: string, patientId: string, auth: string): Promise<CallerContext | null> {
+        try {
+            const [leadData, appointmentsData, callsData, activitiesData] = await Promise.all([
+                this.platform.queryWithAuth<any>(
+                    `{ lead(filter: { id: { eq: "${leadId}" } }) {
+                        id contactName { firstName lastName }
+                        contactPhone { primaryPhoneNumber }
+                        source status interestedService
+                        aiSummary contactAttempts lastContacted
+                        utmCampaign patientId
+                    } }`,
+                    undefined, auth,
+                ),
+                patientId ? this.platform.queryWithAuth<any>(
+                    `{ appointments(first: 10, filter: { patientId: { eq: "${patientId}" } }, orderBy: [{ scheduledAt: DescNullsLast }]) { edges { node {
+                        scheduledAt status doctorName department reasonForVisit
+                    } } } }`,
+                    undefined, auth,
+                ) : Promise.resolve(null),
+                this.platform.queryWithAuth<any>(
+                    `{ calls(first: 10, filter: { leadId: { eq: "${leadId}" } }, orderBy: [{ startedAt: DescNullsLast }]) { edges { node {
+                        startedAt direction durationSec disposition agentName
+                    } } } }`,
+                    undefined, auth,
+                ),
+                this.platform.queryWithAuth<any>(
+                    `{ leadActivities(first: 10, filter: { leadId: { eq: "${leadId}" } }, orderBy: [{ occurredAt: DescNullsLast }]) { edges { node {
+                        activityType summary occurredAt outcome
+                    } } } }`,
+                    undefined, auth,
+                ),
+            ]);
+
+            const lead = leadData?.lead;
+            if (!lead) return null;
+
+            const firstName = lead.contactName?.firstName ?? '';
+            const lastName = lead.contactName?.lastName ?? '';
+
+            return {
+                leadId,
+                patientId: patientId || lead.patientId || '',
+                name: `${firstName} ${lastName}`.trim() || 'Unknown',
+                phone: lead.contactPhone?.primaryPhoneNumber ?? '',
+                isNew: false,
+                leadSource: lead.source ?? null,
+                leadStatus: lead.status ?? null,
+                interestedService: lead.interestedService ?? null,
+                aiSummary: lead.aiSummary ?? null,
+                contactAttempts: lead.contactAttempts ?? 0,
+                lastContacted: lead.lastContacted ?? null,
+                utmCampaign: lead.utmCampaign ?? null,
+                appointments: (appointmentsData?.appointments?.edges ?? []).map((e: any) => e.node),
+                calls: (callsData?.calls?.edges ?? []).map((e: any) => ({
+                    startedAt: e.node.startedAt,
+                    direction: e.node.direction,
+                    duration: e.node.durationSec,
+                    disposition: e.node.disposition,
+                    agentName: e.node.agentName,
+                })),
+                activities: (activitiesData?.leadActivities?.edges ?? []).map((e: any) => e.node),
+            };
+        } catch (err: any) {
+            this.logger.warn(`[CALLER-CTX] Build failed: ${err.message}`);
+            return null;
+        }
+    }
+
+    renderForPrompt(ctx: CallerContext): string {
+        const lines: string[] = [];
+        lines.push(`## CURRENT CALLER: ${ctx.name}`);
+        lines.push(`Phone: ${ctx.phone}`);
+        if (ctx.leadSource) lines.push(`Source: ${ctx.leadSource}`);
+        if (ctx.leadStatus) lines.push(`Status: ${ctx.leadStatus}`);
+        if (ctx.interestedService) lines.push(`Interested in: ${ctx.interestedService}`);
+        if (ctx.utmCampaign) lines.push(`Campaign: ${ctx.utmCampaign}`);
+        if (ctx.contactAttempts > 0) lines.push(`Contact attempts: ${ctx.contactAttempts}`);
+        if (ctx.lastContacted) lines.push(`Last contacted: ${ctx.lastContacted}`);
+
+        if (ctx.aiSummary) {
+            lines.push(`\nAI Summary: ${ctx.aiSummary}`);
+        }
+
+        if (ctx.appointments.length > 0) {
+            lines.push(`\n### Appointments (${ctx.appointments.length})`);
+            for (const a of ctx.appointments) {
+                const date = a.scheduledAt ? new Date(a.scheduledAt).toLocaleDateString('en-IN', { day: 'numeric', month: 'short', year: 'numeric' }) : '?';
+                lines.push(`- ${date} | ${a.doctorName ?? '?'} (${a.department ?? '?'}) | ${a.status}${a.reasonForVisit ? ` | ${a.reasonForVisit}` : ''}`);
+            }
+        } else {
+            lines.push('\nNo appointments on record.');
+        }
+
+        if (ctx.calls.length > 0) {
+            lines.push(`\n### Call History (last ${ctx.calls.length})`);
+            for (const c of ctx.calls) {
+                const date = c.startedAt ? new Date(c.startedAt).toLocaleDateString('en-IN', { day: 'numeric', month: 'short', year: 'numeric' }) : '?';
+                const dur = c.duration ? `${Math.floor(c.duration / 60)}m${c.duration % 60}s` : '?';
+                lines.push(`- ${date} | ${c.direction ?? '?'} | ${dur} | ${c.disposition ?? 'No disposition'}${c.agentName ? ` | Agent: ${c.agentName}` : ''}`);
+            }
+        }
+
+        if (ctx.activities.length > 0) {
+            lines.push(`\n### Recent Activity (last ${ctx.activities.length})`);
+            for (const a of ctx.activities) {
+                const date = a.occurredAt ? new Date(a.occurredAt).toLocaleDateString('en-IN', { day: 'numeric', month: 'short', year: 'numeric' }) : '?';
+                lines.push(`- ${date} | ${a.activityType}${a.summary ? `: ${a.summary}` : ''}${a.outcome ? ` → ${a.outcome}` : ''}`);
+            }
+        }
+
+        return lines.join('\n');
+    }
+}