AI Interview Coordinator for Boutique Recruiting Firms

Automatically schedule interviews across candidates and panelists without the back-and-forth.

voice-agent scheduling recruiting nextjs hono livekit openai google-calendar agent-mesh

The problem

A recruiting coordinator at a 10-person firm juggles 30+ interviews per week, each requiring alignment between a candidate and 3-5 panelists. The constant email ping-pong to find mutual availability consumes 15+ hours weekly. The coordinator needs an agent that can negotiate calendars, send invites, and reschedule conflicts autonomously.

Built from

Intro

This tutorial walks through the AI Interview Coordinator for Boutique Recruiting Firms, a voice-first agent that schedules interviews across candidates and panelists. You’ll build a Next.js 16 application using 6 REAA packages (agent-mesh, agent-mesh-router, agent-handoff, agent-handoff-routing, agent-memory, agent-memory-retrieval) plus Google Calendar, LiveKit for voice, OpenAI for LLM reasoning, and Langfuse for observability. By the end you’ll have a working agent that accepts scheduling requests, routes them to the right sub-agent, checks calendar availability, creates Google Calendar events, and serves everything through REST and MCP endpoints.

Prerequisites

Node.js >= 22 and pnpm >= 10 installed
A Google Cloud service account with Calendar API enabled (for calendar operations)
A LiveKit server or cloud account (for voice agent rooms)
An OpenAI API key with access to gpt-4o, text-embedding-3-small, and whisper-1/tts-1
(Optional) A Langfuse account for LLM observability
Familiarity with TypeScript, Next.js App Router, and basic Zod schemas

Step 1: Scaffold the project

Create a Next.js 16 project with the App Router, all dependencies exact-pinned in package.json, a tsconfig.json with strict: true and NodeNext module resolution, vitest.config.ts with 90% coverage thresholds, and eslint.config.mjs with flat config.

Your package.json should look like this:

Example artifact

A complete, working implementation of this recipe — downloadable as a zip or browsable file by file. Generated by our build pipeline; tested with full coverage before publishing.

Download example (zip)Browse files

213 kB·117 tests·97.3% coverage·vitest passing

SHA-256edf16857214790bf05d48f2637e5a3d3dfadaeae4103d8c8eba78dfa4273e98e

Book a conversation All solutions

Comments

Loading comments…

Intro

Prerequisites

Node.js >= 22 and pnpm >= 10 installed
A Google Cloud service account with Calendar API enabled (for calendar operations)
A LiveKit server or cloud account (for voice agent rooms)
An OpenAI API key with access to gpt-4o, text-embedding-3-small, and whisper-1/tts-1
(Optional) A Langfuse account for LLM observability
Familiarity with TypeScript, Next.js App Router, and basic Zod schemas

Step 1: Scaffold the project

Your package.json should look like this:

import { CapabilityBasedRouter, AgentRegistry } from "@reaatech/agent-handoff-routing"; import { createHandoffConfig, pickDefined, HandoffError, TypedEventEmitter, withRetry } from "@reaatech/agent-handoff"; import type { HandoffPayload, RoutingDecision } from "@reaatech/agent-handoff"; export type { HandoffError, TypedEventEmitter, withRetry }; export class InterviewAgentRouter { private readonly router: CapabilityBasedRouter; private readonly registry: AgentRegistry; private readonly handoffConfig: ReturnType<typeof createHandoffConfig>; constructor() { const handoffConfig = createHandoffConfig({ routing: { minConfidenceThreshold: 0.6 } }); this.handoffConfig = handoffConfig; pickDefined({ handoffConfig }); this.router = new CapabilityBasedRouter({ minConfidenceThreshold: 0.7, ambiguityThreshold: 0.15, maxAlternatives: 3, policy: "best_effort", }); this.registry = new AgentRegistry(); this.registry.register({ agentId: "scheduler", agentName: "Scheduler Agent", skills: ["calendar", "availability", "timezone", "scheduling"], domains: ["hr", "recruiting", "logistics"], maxConcurrentSessions: 5, currentLoad: 0, languages: ["en"], specializations: [], availability: "available", version: "1.0.0", }); this.registry.register({ agentId: "notifier", agentName: "Notifier Agent", skills: ["email", "notification", "invitation", "messaging"], domains: ["hr", "communication"], maxConcurrentSessions: 5, currentLoad: 0, languages: ["en"], specializations: [], availability: "available", version: "1.0.0", }); this.registry.register({ agentId: "followup", agentName: "Follow-up Agent", skills: ["rescheduling", "negotiation", "conflict-resolution", "calendar"], domains: ["hr", "recruiting"], maxConcurrentSessions: 5, currentLoad: 0, languages: ["en"], specializations: [], availability: "available", version: "1.0.0", }); } async routeToAgent(payload: HandoffPayload): Promise<RoutingDecision> { return this.router.route(payload, this.registry.getAll()); } } export const agentRouter = new InterviewAgentRouter();

import { AgentMemory, OpenAILLMProvider, MemoryType, OpenAIEmbeddingProvider, } from "@reaatech/agent-memory"; import { MemoryRetriever, ContextInjector, RetrievalStrategy, } from "@reaatech/agent-memory-retrieval"; import { config } from "../lib/config.js"; import type { ConversationTurn } from "../lib/interview-types.js"; export class InterviewMemoryService { private memory: AgentMemory; private retriever: MemoryRetriever; private injector: ContextInjector; constructor() { this.memory = new AgentMemory({ storage: { provider: "memory" }, embedding: { provider: "openai", model: "text-embedding-3-small", apiKey: config.OPENAI_API_KEY, }, extraction: { llmProvider: new OpenAILLMProvider({ apiKey: config.OPENAI_API_KEY, model: "gpt-4o-mini", }), enabledTypes: [MemoryType.FACT, MemoryType.PREFERENCE], batchSize: 10, confidenceThreshold: 0.7, }, }); const embeddingProvider = new OpenAIEmbeddingProvider({ apiKey: config.OPENAI_API_KEY, model: "text-embedding-3-small", }); this.retriever = new MemoryRetriever( this.memory.getStorage(), embeddingProvider, { defaultLimit: 5, useCrossEncoder: false, diversityFactor: 0.3, strategies: [ RetrievalStrategy.SEMANTIC, RetrievalStrategy.RECENCY, ], }, ); this.injector = new ContextInjector(); } async storeCandidatePreference( candidateId: string, conversation: ConversationTurn[], ): Promise<void> { const mapped = conversation.map((c) => ({ speaker: c.speaker as "user" | "agent", content: c.content, timestamp: c.timestamp, })); await this.memory.extractAndStore(mapped); } async retrieveCandidateContext( candidateId: string, query: string, ): Promise<string> { const memories = await this.retriever.retrieve(query, { limit: 5, tenantId: candidateId, }); return this.injector.injectMemoriesIntoContext([], memories, 2000); } async storeInterviewMemory( interviewId: string, turn: { role: string; content: string }, ): Promise<void> { const conversationTurn = { speaker: turn.role as "user" | "agent", content: turn.content, timestamp: new Date(), }; await this.memory.extractAndStore([conversationTurn]); } async runMaintenance(tenantId?: string): Promise<void> { await this.memory.runMaintenance(tenantId); } async dispose(): Promise<void> { await this.memory.close(); } } export const memoryService = new InterviewMemoryService();

import { Hono } from "hono"; import { ContextPacketSchema, type ContextPacket, } from "@reaatech/agent-mesh"; import { z } from "zod"; import { coordinator } from "./interview-coordinator.js"; import { ScheduleRequestSchema, AvailabilityQuerySchema, type ScheduleRequest, type AvailabilityQuery, type AgentResponse, } from "../lib/interview-types.js"; const honoApp = new Hono(); honoApp.post("/mcp/schedule", async (c) => { try { const body: unknown = await c.req.json(); const packet: ContextPacket = ContextPacketSchema.parse(body); const scheduleRequest: ScheduleRequest = ScheduleRequestSchema.parse({ candidateId: packet.entities.candidateId ?? packet.employee_id, panelistIds: packet.entities.panelistIds ?? [], durationMinutes: packet.entities.durationMinutes ?? 60, dateRange: packet.entities.dateRange ?? { start: new Date().toISOString(), end: new Date(Date.now() + 7 * 86400000).toISOString(), }, notes: packet.raw_input, }); const result: AgentResponse = await coordinator.scheduleInterview( scheduleRequest, ); return c.json(result); } catch (error) { if (error instanceof z.ZodError) { return c.json({ error: error.issues }, 400); } throw error; } }); honoApp.post("/mcp/availability", async (c) => { try { const body: unknown = await c.req.json(); const packet: ContextPacket = ContextPacketSchema.parse(body); const query: AvailabilityQuery = AvailabilityQuerySchema.parse({ candidateId: packet.entities.candidateId ?? packet.employee_id, panelistIds: packet.entities.panelistIds ?? [], dateRange: packet.entities.dateRange ?? { start: new Date().toISOString(), end: new Date(Date.now() + 7 * 86400000).toISOString(), }, }); const slots = await coordinator.checkAvailability( query.candidateId, query.panelistIds, query.dateRange, ); return c.json({ slots }); } catch (error) { if (error instanceof z.ZodError) { return c.json({ error: error.issues }, 400); } throw error; } }); export { honoApp }; export default honoApp;

import { NextRequest, NextResponse } from "next/server"; import { z } from "zod"; import { coordinator } from "../../../../src/services/interview-coordinator.js"; import { InterviewNotFoundError } from "../../../../src/lib/errors.js"; import { TimeSlotSchema } from "../../../../src/lib/interview-types.js"; const RescheduleBodySchema = z.object({ newSlot: TimeSlotSchema }); const CancelBodySchema = z.object({ reason: z.string() }); export async function GET( req: NextRequest, { params }: { params: Promise<{ id: string }> }, ): Promise<NextResponse> { const { id } = await params; try { const interview = coordinator.getInterview(id); return NextResponse.json(interview); } catch (error) { if (error instanceof InterviewNotFoundError) { return NextResponse.json( { error: error.message }, { status: 404 }, ); } return NextResponse.json( { error: "Internal server error" }, { status: 500 }, ); } } export async function PATCH( req: NextRequest, { params }: { params: Promise<{ id: string }> }, ): Promise<NextResponse> { const { id } = await params; try { const body = RescheduleBodySchema.parse(await req.json()); const response = await coordinator.rescheduleInterview(id, body.newSlot); return NextResponse.json(response); } catch (error) { if (error instanceof InterviewNotFoundError) { return NextResponse.json( { error: error.message }, { status: 404 }, ); } return NextResponse.json( { error: "Internal server error" }, { status: 500 }, ); } } export async function DELETE( req: NextRequest, { params }: { params: Promise<{ id: string }> }, ): Promise<NextResponse> { const { id } = await params; try { const body = CancelBodySchema.parse(await req.json()); await coordinator.cancelInterview(id, body.reason); return NextResponse.json({ status: "cancelled" }, { status: 200 }); } catch (error) { if (error instanceof InterviewNotFoundError) { return NextResponse.json( { error: error.message }, { status: 404 }, ); } return NextResponse.json( { error: "Internal server error" }, { status: 500 }, ); } }

AI Interview Coordinator for Boutique Recruiting Firms

The problem

Built from

Intro

Prerequisites

Step 1: Scaffold the project

Example artifact

Comments

Intro

Prerequisites

Step 1: Scaffold the project

Step 2: Set up environment variables and config

Step 3: Define domain types with Zod

Step 4: Create typed error classes

Step 5: Build the Calendar adapter

Step 6: Set up the Agent Router with capability-based routing

Step 7: Build the Memory Service for persistent context

Step 8: Build the Interview Coordinator — the orchestration core

Step 9: Create the LiveKit voice agent

Step 10: Add the Hono MCP server

Step 11: Wire up the Next.js API routes

Step 12: Add the instrumentation hook for startup initialization

Step 13: Run the tests

Next steps