Home About Projects Case Studies Research Blog Open Source Now

Switch Theme

AI Applications🤖 AI Applications

CareerAI

An AI-powered career coach that analyzes resumes, provides ATS feedback, generates personalized learning roadmaps, and prepares users for interviews using Google Gemini AI.

View GitHub Code

Benchmark Results & Metrics

ATS Score Increase

+35%

Feedback Latency

<1.5s

Resume Match Rate

94%

Gemini LLM

1.5 Pro

01 // Overview

CareerAI is a comprehensive, end-to-end AI career coaching platform. It integrates resume parsing, ATS compatibility checking, automated learning path generation, and real-time interactive interview prep powered by Google Gemini.

02 // The Problem

Standard ATS systems reject over 70% of qualified resumes due to formatting and keyword mismatch, while candidates struggle to find personalized study resources or practice realistic interviews.

03 // System Architecture

Agentic Resume Parsing & Assessment Pipeline

Resumes are parsed using custom extractors, then sent to a Gemini assessment chain that generates structured ATS feedback and matching roadmaps. Mock interviews run on a stateful chat session.

ARCHITECTURAL DATA FLOW:Resume PDF -> PDF Extractor -> Gemini ATS Chain -> JSON Feedback -> Roadmap Generator -> Frontend UI

Architecture Components:

Resume PDF Parser

Gemini ATS Evaluator

Roadmap Generation Service

Interactive Mock Interview Engine

04 // Implementation

Built the frontend using React and TypeScript. The backend runs on FastAPI with Pydantic schemas, interacting with Gemini AI APIs using LangChain for structural output validation.

05 // Execution Workflow

11. User uploads resume PDF.

22. System extracts text and analyzes keyword density.

33. Gemini generates ATS score and improvements.

44. Personalized learning roadmap is rendered.

55. Interactive mock interview session starts.

06 // Technology Stack

Frontend

ReactTypeScriptTailwind CSS

Backend & AI

FastAPIGemini AILangChain

07 // Technical Challenges

⚠️Parsing multi-column PDF resumes accurately without losing layout context.

⚠️Minimizing token usage when sending large resume texts to LLMs for ATS scoring.

Lessons Learned

•Using structured JSON schemas with Gemini ensures deterministic scoring outputs.
•Pre-filtering PDF styling reduces token overhead and increases parse reliability.

Future Improvements

•Adding voice-based interactive mock interviews using WebRTC and Gemini Multimodal Live API.
•Integrating direct job application portal sync.