Nexa – AI-Powered Interview Preparation Platform

Prepare Smarter. Get Hired Faster.

Comment 1

Inspiration

Preparing for technical interviews today is unnecessarily fragmented. Candidates have to jump between job boards, resume reviewers, YouTube, blogs, documentation, and random ChatGPT prompts—often without a clear, structured plan tailored to the specific job they are applying for.

As students and early-career developers ourselves, we experienced this problem firsthand. Even when you find a good job posting, figuring out what skills you are missing, what to study, where to study from, and how to practice is a time consuming and chaotic process.

We wanted to build something that feels like a personal AI mentor: Something that understands your resume, understands the job, builds a study plan automatically, and then trains you for that specific role.

That idea became Nexa.

What it does

Nexa is an end-to-end AI-powered interview preparation platform.

It:

  • Parses your resume using AI
  • Finds relevant jobs automatically
  • Analyzes job requirements and identifies missing skills
  • Searches and scrapes the best learning resources from the web
  • Summarizes and converts them into a personalized knowledge base
  • Builds a RAG (Retrieval-Augmented Generation) system for that specific job
  • Lets you chat with an AI interview coach that knows:
    • Your resume
    • The job role
    • The company
    • The required skills
    • The curated learning material
  • Two modes:
    • Interview and preparation
    • One mode teaches you while the other helps you practice

In short: You click “Start Preparing” and Nexa builds your entire interview preparation pipeline automatically.

How we built it

We built Nexa as a full-stack AI system with a real production-style architecture.

AI & Data Pipeline

  • Groq (Llama 3.3 70B) Resume parsing and skill extraction
  • SerpAPI Job discovery
  • Tavily AI Learning resource discovery
  • Web Scraper Extracts raw content from articles and docs
  • GPT-4o (GitHub Models) Summarizes content into interview-focused material
  • Chunking + Embeddings (HuggingFace all-mpnet-base-v2) Builds vector representations
  • Vector Search (Cosine Similarity) Retrieves relevant knowledge
  • GPT-4o → Generates contextual interview answers using RAG
  • Instead of a vector store like Chroma DB or FAISS, we used Memory Vector Store to make serverless RAG possible.

Tech Stack

  • Frontend: Next.js 15, TypeScript, Tailwind, shadcn/ui, Framer Motion
  • Backend: Node.js API routes, Supabase Auth
  • RAG System: Custom ingestion + retrieval pipeline
  • Auth & Storage: Supabase
  • Deployment: Vercel

System Flow

Resume → Job Search → Skill Gap Analysis → Resource Discovery → Scraping → Summarization → Chunking → Embedding → Vector DB → AI Coach Chat

Challenges we ran into

  • Automating the entire pipeline reliably without manual steps
  • Serverless deployment being a constraint due to zero budget
  • Handling very noisy web data during scraping and making summaries actually useful
  • Designing a RAG system that stays fast while embedding large documents
  • Making sure the AI responses stay grounded in retrieved context
  • Managing API limits and latency across multiple AI services
  • Orchestrating multiple asynchronous workflows in a clean and debuggable way
  • Ensuring the UX feels instant and magical despite heavy background processing

Accomplishments that we're proud of

  • Built a fully automated, zero-manual-step interview preparation pipeline
  • Implemented a real production-style RAG system, not a demo
  • Created job-specific, company-specific AI coaching sessions
  • Integrated multiple AI providers into one coherent system
  • Designed a clean, futuristic UI with a strong product identity
  • Achieved persistent per-job chat history and knowledge bases
  • Turned a complex multi-stage AI workflow into a single-button experience

What we learned

  • How to design and build real-world RAG systems
  • How to orchestrate multi-stage AI pipelines
  • How to work with LLMs as infrastructure, not just chatbots
  • The importance of data quality before embeddings
  • How to design AI products that feel useful, not gimmicky
  • How to balance UX, performance, and system complexity
  • How real AI products require engineering first, prompting second

What's next for Nexa – AI-Powered Interview Preparation Platform

We want to evolve Nexa into a complete AI career co-pilot:

  • Mock interview mode with voice and real-time feedback
  • Skill progression tracking and readiness scoring
  • Adaptive learning paths per user
  • Company-specific interview pattern analysis
  • Multi-role preparation support
  • Scalable vector storage and enterprise-grade RAG
  • Multi-agent AI coaching (HR interviewer, technical interviewer, behavioral coach)

Our long-term vision: Nexa should feel like having a personal AI career mentor that stays with you throughout your professional journey.

Built With

Share this project:

Updates