CareerCopilot AI

Stop browsing and start applying. CareerCopilot AI automates the tedious parts of your job search by finding targets across multiple boards and optimizing your application workflow.

Comment

CareerCopilot AI

About the Project

CareerCopilot AI is a comprehensive, AI-powered career assistant designed to democratize access to premium career coaching and automation tools. In today's hyper-competitive job market, candidates often struggle with three key bottlenecks: manual resume tailoring, fragmented job searching across multiple platforms, and lack of personalized interview preparation. CareerCopilot solves this by acting as an intelligent agent that navigates the entire lifecycle of a job application.

Inspiration

The inspiration behind CareerCopilot came from observing the inefficiencies in the traditional job hunt. Highly skilled candidates often get rejected by Applicant Tracking Systems (ATS) simply because their resumes aren't optimized for specific keywords, or they miss out on opportunities because they can't scan every job board daily. We wanted to build a "Jarvis for your Career"—a tool that doesn't just give generic advice but actually performs the heavy lifting: parsing documents, scraping the web for live opportunities, and generating high-conversion application assets tailored to specific roles.

How We Built It

The application is built as a modern, responsive Single Page Application (SPA) using React and TypeScript, styled with Tailwind CSS for a clean, "glassmorphism" aesthetic.

Core Architecture:

  1. AI Engine: We leverage Google's Gemini 3 Flash model via the @google/genai SDK. This model powers:

    • Resume Parsing: Converting unstructured PDF text into strict JSON profiles.
    • Content Generation: Creating ATS-friendly HTML resumes, cover letters, and outreach emails.
    • Live Simulation: We utilized the Gemini Live API to build a real-time voice interview simulator that speaks back to the candidate with low latency.

  2. Job Aggregation Layer: Instead of relying on expensive third-party APIs, we built a custom client-side scraper (using CORS proxies) that aggregates real-time listings from major Indian job portals like Naukri, Hirist, and FoundIt.

  3. Document Processing: We integrated pdfjs-dist to perform client-side OCR and text extraction from PDF resumes, ensuring user data privacy by processing files in the browser before sending text to the AI.

Challenges We Faced

  • Structuring Unstructured Data: One of the biggest hurdles was getting the AI to consistently extract structured JSON (Skills, Experience dates) from messy, diverse resume formats. We overcame this with rigorous prompt engineering and schema validation.
  • Real-time Audio Processing: Implementing the Gemini Live API in the browser was technically demanding. We had to manually handle PCM audio encoding/decoding, buffer management for smooth playback, and synchronization between user input and model output to prevent "talking over" each other.
  • Cross-Origin Resource Sharing (CORS): Fetching job data directly from the frontend triggered browser security blocks. We implemented a robust proxying strategy to bypass these restrictions while normalizing the disparate data structures from different job boards into a unified Job interface.

What We Learned

Building CareerCopilot taught us the immense potential of Multimodal AI. Moving beyond simple text chatbots to an agent that can "read" files, "see" job descriptions, and "speak" in an interview setting creates a far more immersive and helpful user experience. We also learned the importance of "Human-in-the-loop" design—allowing users to verify the AI's extraction (in the Profile Verification step) built trust and improved the accuracy of downstream tasks.

Built With

Share this project:

Updates