Rakib Hassan - Full-Stack Developer
Projects
Blog
About
Contact
RESUMEGITHUB
Back to ProjectsProjectsFull Stack / AI SaaS

MockMateAI-IntelligentInterviewCoach

A voice-first AI interview coach that replicates real technical hiring panels — analyzing verbal responses, generating dynamic follow-ups, and delivering actionable rubric scoring under 1.5s latency with Google Gemini.

Client

Independent Build

Year

2026

Role

Lead Full Stack Developer

LiveSource
MockMate AI - Intelligent Interview Coach — case study by Rakib Hassan, Full-Stack Developer
0%

Server Side Logic

0s

AI Latency

0/100

Performance Score

The Challenge

The main challenge was handling real-time browser-native speech recognition while ensuring seamless synchronization with the AI backend. Additionally, minimizing latency for AI responses and managing strict type safety across the new Next.js 16 Server Actions ecosystem was critical.

Technologies used

Frontend
(5)
Next.js 16React 19TypeScriptTailwind CSS v4Shadcn UI
Backend
(5)
Server ActionsClerk AuthGoogle Gemini AISpeech APIZod
Database
(2)
MongoDBMongoose
DevOps
(1)
Vercel

The Solution

→ Next.js 16 Server Actions + MongoDB for direct, zero-API data mutations. → Browser Speech API + Gemini 1.5 Flash for rapid, real-time AI feedback. → Tailwind CSS v4 + Shadcn UI for a clean, modern interface.

typescript.ts
Loading code...

Key Decisions

Why Server Actions over API Routes?

Since we needed direct, low-latency database mutations from the client form, Server Actions eliminated the need to maintain boilerplate API routes.

Why Gemini 1.5 Flash?

Interview feedback requires near-instant responses. Gemini Flash offered the perfect balance of speed and high reasoning capabilities over heavier models.

Why native Speech Recognition?

Instead of dealing with heavy audio file uploads and server-side transcription, using the browser API handled it locally with zero cost and zero latency.

What I Learned

Hardest Part

Handling edge cases with browser compatibility for the native Speech Recognition API, especially across different mobile devices.

Biggest Insight

Prompt engineering is essentially programming in English. Crafting the strict JSON schema prompt for the AI was the most critical piece of the architecture.

Do Differently

I would implement WebSockets for streaming the AI response word-by-word instead of waiting for the full JSON block to load.

Key Highlights

AI Question Generation

Generates 5 unique technical questions based on Job Role & Tech Stack.

Voice-to-Text Engine

Records user answers in real-time using browser-native Speech API.

Instant AI Feedback

Analyzes answers to provide 1-10 ratings and improvement tips.

Interview History

Dashboard to track performance trends and review past feedback.

Next project · E-Commerce Solution

Input Gears — Tech Gadget Store

Next case study