Hexafusion Technical Assessment

Role: Full-Stack AI Developer – Voice, Chatbot & Automation Platforms

Submission Deadline: Friday, May 16, 2025 – 11:59 PM PT
Submit to: [email protected]

Overview

This assessment is designed to evaluate your ability to build intelligent, production-grade AI systems that go beyond basic chatbots. You will demonstrate your skills in AI agent orchestration, automation workflows, voice interaction design, persistent memory, and secure backend engineering.

• You must complete 2 challenges, each in under 2 - 4 hours of coding time

• You may host them privately and provide a link, or send a recording and code repo

• No need to purchase Twilio credits – alternatives are provided

Challenge 1: Voice AI Agent with Memory and Caller Recognition

Objective
Build a fully interactive voice agent that answers calls, converses naturally, collects caller information, stores it in a database, and recognizes returning callers.

Requirements

1. Conversation Flow

   • Greet the caller

   • Ask for: full name, company name, and reason for calling

   • Use natural language understanding (OpenAI or equivalent) to handle variations

   • Support multi-turn conversation and fallback handling

2. Voice Integration

   • Use Amazon Polly (Matthew) for text-to-speech

   • Use Twilio Voice (trial account allowed) OR a browser-based simulation using Web Speech API

   • Use Twilio or Whisper for speech-to-text

3. Data Persistence

   • Store caller data (name, phone, company, reason, timestamp) in MongoDB or PostgreSQL

   • When the same number calls again, greet them by name and recall their last interaction

4. Security and Configuration

   • Use .env to manage all API keys and secrets

   • Validate inputs, handle errors, and log activity

Deliverables

• Hosted demo (Twilio, browser simulation, or video walkthrough)

• Source code in GitHub or zipped folder

• Short (3–5 min) video walkthrough showing:

  • The voice interaction

  • How memory and user recognition work

  • Code explanation

Notes on Free Hosting

• Twilio’s free trial allows calling verified personal numbers and provides enough credit for testing

• If you cannot use Twilio, simulate the flow with a browser-based voice UI (e.g., HTML + JavaScript + Polly audio + OpenAI)

Challenge 2: End-to-End AI Workflow with Automation and Reporting

Objective
Create a modular AI workflow that processes incoming data, performs a smart action, logs the results, and displays them in a dashboard.

Requirements

1. Trigger Input

   • Accept input via a web form, email, or webhook (manual POST request is acceptable)

2. AI Processing

   • Use OpenAI to analyze or summarize the input

   • Submit your custom prompt and explain your logic

3. Automation

   • Take action based on AI output:

     • Send a notification (e.g., Slack, Teams, or email)

     • Store in Notion, Google Sheets, or a database

4. Dashboard

   • Display at least the last 10 processed items in a table or log view

   • Include timestamp, original input, and AI-generated output

   • Optional: add basic charts or summaries

5. Workflow Engine

   • Use Zapier, n8n, or custom-built logic (Node.js, Python, etc.)

6. Security

   • Use environment variables

   • Avoid exposing any credentials

Deliverables

• Hosted form, dashboard link, or JSON export of workflow

• GitHub repo or zipped folder with source

• 3–5 minute video explaining the architecture, prompt, and demo