AI Career Copilot

A Production-Grade Agentic AI Assistant for Job Search and Career Development

AI Career Copilot is a Telegram-based intelligent system built to streamline the end-to-end job hunting process. It combines semantic understanding, hybrid search, and multi-agent AI orchestration to deliver highly relevant job opportunities and personalized resume optimization.

Project Overview

This project transforms a basic chatbot into a complete agentic AI system capable of:

• Ingesting and structuring resumes from PDF documents
• Performing intelligent job discovery using hybrid lexical and vector search
• Running a multi-agent critique loop to iteratively tailor resumes for specific roles
• Analyzing skill gaps against the current job market
• Delivering personalized technology briefings

The system is designed with production-grade architecture, emphasizing reliability, observability, and cost efficiency.

Core Capabilities

Semantic Onboarding

• Parses PDF resumes in memory using pdfplumber
• Extracts structured profile data using Gemini + Instructor with strict Pydantic validation
• Generates 768-dimensional embeddings and stores profiles in PostgreSQL with pgvector

Two-Stage Hybrid Job Search

• Combines BM25 full-text search with semantic vector similarity for fast retrieval
• LLM-powered reranking with detailed reasoning and skill-gap analysis
• India-focused job ingestion from platforms including Instahyre, Cutshort, Hirist, and public APIs

Multi-Agent Resume Tailoring

• Built using LangGraph, featuring two specialized agents:
  • Writer Agent: Crafts tailored resume sections based on target job descriptions
  • ATS Critic Agent: Evaluates alignment (0-100 score) and provides actionable feedback
• Iterative critique loop continues until the resume meets high standards or maximum iterations are reached
• Final output compiled into clean, professional PDFs using Typst

Additional Intelligence

• Skill gap analysis comparing user profile against active job market requirements
• Personalized weekly learning roadmaps
• Automated tech news aggregation and summarization

Production Features

• Full async architecture with FastAPI
• Redis caching for performance and cost optimization
• Comprehensive observability using OpenTelemetry and LangSmith
• Robust error handling and multi-tenancy isolation

Technical Architecture

• Backend: FastAPI (async), SQLAlchemy + asyncpg
• Agent Framework: LangGraph (StateGraph with conditional routing)
• AI Layer: Gemini 1.5 Flash + Instructor for structured outputs
• Vector Database: PostgreSQL 16 + pgvector
• Document Engine: Typst for high-quality PDF generation
• Scraping: Playwright (responsible, rate-limited usage)
• Monitoring: OpenTelemetry, LangSmith

Key Learnings & Highlights

This project served as a deep dive into building agentic AI systems. The multi-agent resume tailoring workflow demonstrates practical implementation of autonomous AI collaboration, state management, and iterative reasoning — core concepts in modern agentic AI development.

The system reflects production thinking through proper caching strategies, observability, async design, and cost-aware LLM usage.

---