Creating a project workflow using a voice assistant and converting it into an approval card.

apps/api/src/ai/ai-ask.service.ts

@@ -1,8 +1,12 @@
 import { Injectable, NotFoundException } from '@nestjs/common';
 import type {
+  ActivityType,
   AskMentoResponse,
   AskMentoSourceDto,
   PersonAiSummaryDto,
+  VoiceActivityDraft,
+  VoiceCommandDraft,
+  VoicePersonDraft,
 } from '@mento/shared';
 import { PrismaService } from '../prisma/prisma.service';
 import {
@@ -674,4 +678,155 @@ Rules:
 
     return { context: lines.join('\n').slice(0, 14000), catalog };
   }
+
+  async parseVoiceCommand(actor: Actor, text: string): Promise<VoiceCommandDraft> {
+    const now = new Date();
+    const today = new Intl.DateTimeFormat('en-CA', {
+      timeZone: 'Asia/Tehran',
+      year: 'numeric',
+      month: '2-digit',
+      day: '2-digit',
+    }).format(now);
+
+    const raw = await this.ai.completeJson<{
+      intent?: string;
+      summary?: string;
+      person?: Partial<VoicePersonDraft> | null;
+      activity?: Partial<VoiceActivityDraft> | null;
+    }>(
+      `You convert one Persian voice command into a CRM action draft.
+Return ONLY JSON:
+{
+  "intent": "create_person" | "create_activity" | "none",
+  "summary": string,
+  "person": { "firstName": string, "lastName": string, "mobile": string, "phone": string, "email": string, "website": string, "referrer": string, "position": string, "notes": string } | null,
+  "activity": { "title": string, "type": "call"|"meeting"|"task"|"note", "description": string, "dueAt": string|null, "personName": string } | null
+}
+Rules:
+- create_person only when the user asks to add/create a customer or contact.
+- create_activity only when the user asks to add a follow-up, call, meeting, or task.
+- Otherwise intent is none. Do not invent a record.
+- Do not invent phone numbers, emails, or names that were not spoken.
+- Mobile/phone: digits only, Iranian mobile as 09xxxxxxxxx.
+- dueAt: ISO 8601 in Asia/Tehran if a time is mentioned, else null. Today is ${today}.
+- summary: one short Persian sentence describing what will be saved.
+- Unused string fields are empty strings.`,
+      text.trim(),
+    );
+
+    const intent =
+      raw.intent === 'create_person' || raw.intent === 'create_activity'
+        ? raw.intent
+        : 'none';
+
+    if (intent === 'create_person') {
+      const person = cleanPerson(raw.person);
+      const missing: string[] = [];
+      if (!person.firstName) missing.push('نام');
+      return {
+        intent,
+        summary: (raw.summary ?? '').trim() || `ساخت مشتری ${person.firstName} ${person.lastName}`.trim(),
+        missing,
+        canConfirm: missing.length === 0,
+        person,
+        activity: null,
+      };
+    }
+
+    if (intent === 'create_activity') {
+      const activity = cleanActivity(raw.activity);
+      if (activity.personName) {
+        activity.personId = await this.matchPersonId(actor, activity.personName);
+      }
+      const missing: string[] = [];
+      if (!activity.title) missing.push('عنوان فعالیت');
+      return {
+        intent,
+        summary: (raw.summary ?? '').trim() || `ثبت فعالیت «${activity.title}»`,
+        missing,
+        canConfirm: missing.length === 0,
+        person: null,
+        activity,
+      };
+    }
+
+    return {
+      intent: 'none',
+      summary:
+        (raw.summary ?? '').trim() ||
+        'در این مرحله فقط ساخت مشتری یا ثبت فعالیت از روی صوت انجام می‌شود.',
+      missing: [],
+      canConfirm: false,
+      person: null,
+      activity: null,
+    };
+  }
+
+  private async matchPersonId(actor: Actor, name: string): Promise<string | null> {
+    const q = name.trim();
+    if (q.length < 2) return null;
+    const rows = await this.prisma.person.findMany({
+      where: {
+        organizationId: actor.organizationId,
+        ...ownerWhere(actor),
+        OR: [
+          { firstName: { contains: q, mode: 'insensitive' } },
+          { lastName: { contains: q, mode: 'insensitive' } },
+        ],
+      },
+      select: { id: true, firstName: true, lastName: true },
+      take: 5,
+    });
+    const exact = rows.filter((row) => `${row.firstName} ${row.lastName}`.trim() === q);
+    if (exact.length === 1) return exact[0]!.id;
+    if (rows.length === 1) return rows[0]!.id;
+    return null;
+  }
+}
+
+function cleanText(value: unknown, max = 200) {
+  return String(value ?? '').replace(/\s+/g, ' ').trim().slice(0, max);
+}
+
+function digitsOnly(value: unknown) {
+  const fa = '۰۱۲۳۴۵۶۷۸۹';
+  const ar = '٠١٢٣٤٥٦٧٨٩';
+  const raw = String(value ?? '')
+    .replace(/[۰-۹]/g, (ch) => String(fa.indexOf(ch)))
+    .replace(/[٠-٩]/g, (ch) => String(ar.indexOf(ch)));
+  let digits = raw.replace(/[^\d+]/g, '');
+  if (digits.startsWith('+98')) digits = `0${digits.slice(3)}`;
+  else if (digits.startsWith('98') && digits.length === 12) digits = `0${digits.slice(2)}`;
+  return digits.slice(0, 20);
+}
+
+function cleanPerson(input: Partial<VoicePersonDraft> | null | undefined): VoicePersonDraft {
+  return {
+    firstName: cleanText(input?.firstName, 80),
+    lastName: cleanText(input?.lastName, 80),
+    mobile: digitsOnly(input?.mobile),
+    phone: digitsOnly(input?.phone),
+    email: cleanText(input?.email, 120),
+    website: cleanText(input?.website, 300),
+    referrer: cleanText(input?.referrer, 120),
+    position: cleanText(input?.position, 120),
+    notes: cleanText(input?.notes, 500),
+  };
+}
+
+const ACTIVITY_TYPES = new Set<ActivityType>(['call', 'meeting', 'task', 'note']);
+
+function cleanActivity(input: Partial<VoiceActivityDraft> | null | undefined): VoiceActivityDraft {
+  const type = ACTIVITY_TYPES.has(input?.type as ActivityType)
+    ? (input?.type as ActivityType)
+    : 'task';
+  const due = cleanText(input?.dueAt, 40);
+  return {
+    title: cleanText(input?.title, 160),
+    type,
+    description: cleanText(input?.description, 500),
+    dueAt: due && !Number.isNaN(Date.parse(due)) ? new Date(due).toISOString() : null,
+    personName: cleanText(input?.personName, 120),
+    personId: null,
+  };
 }

apps/api/src/ai/ai.controller.ts

@@ -3,7 +3,7 @@ import { JwtAuthGuard } from '../auth/jwt-auth.guard';
 import { CurrentUser } from '../auth/current-user.decorator';
 import type { Actor } from '../common/record-scope';
 import { AiAskService } from './ai-ask.service';
-import { AskMentoDto } from './dto';
+import { AskMentoDto, VoiceCommandDto } from './dto';
 import { AiService } from './ai.service';
 
 @Controller('ai')
@@ -19,6 +19,11 @@ export class AiController {
     return { enabled: this.ai.isEnabled() };
   }
 
+  @Post('voice-command')
+  voiceCommand(@CurrentUser() user: Actor, @Body() dto: VoiceCommandDto) {
+    return this.ask.parseVoiceCommand(user, dto.text);
+  }
+
   @Post('ask')
   askMento(@CurrentUser() user: Actor, @Body() dto: AskMentoDto) {
     return this.ask.ask(user, dto.question, dto.personId);

apps/api/src/ai/dto.ts

@@ -1,5 +1,12 @@
 import { IsOptional, IsString, MaxLength, MinLength } from 'class-validator';
 
+export class VoiceCommandDto {
+  @IsString()
+  @MinLength(2)
+  @MaxLength(800)
+  text!: string;
+}
+
 export class AskMentoDto {
   @IsString()
   @MinLength(2)