01
Context & Problem
The modern job application process is severely broken. Tailoring resumes to bypass rigid Applicant Tracking Systems (ATS) requires manual, tedious keyword optimization for every single application.
Simultaneously, tracking application statuses, monitoring incoming recruiter emails, and scheduling interviews creates a chaotic administrative overhead that distracts candidates from actual interview preparation.
The Objective
Build a closed-loop, autonomous system that acts as a cognitive assistant—scraping requirements, evaluating candidate fit, rewriting resumes precisely, and tracking correspondence without manual intervention.
02
Engineering Constraints
-
Hallucination Risk When an LLM rewrites a resume, it must remain 100% faithful to the user's actual history. It cannot invent skills to match a job description.
-
Workflow Orchestration Scraping websites, hitting LLM endpoints, and writing to Google Sheets required robust async handling and error fallbacks to prevent pipeline crashes.
03
Architecture
I adopted a multi-agent orchestration approach using LangGraph to cleanly separate concerns. Rather than one massive prompt, the system relies on specialized nodes.
Agent Nexus
A stateful LangGraph orchestrator that uses a ReAct framework to autonomously query Gmail and Google Sheets for real-time job application tracking and interview management.
Agent Nexus
A high-throughput workflow that automates LinkedIn scraping, filters roles via a Gemini relevance checker, and executes a multi-agent chain to generate tailored, ATS-optimized PDF resumes.
The frontend was built with HTML/Tailwind to deliver a sleek, glassmorphism UI. A Python FastAPI backend bridges the UI to the LangGraph execution engine, allowing real-time chat interactions with the data via Google APIs (Sheets & Gmail).
04
Impact & Value
Time Saved
Reduced manual application preparation time from ~45 minutes to under 2 minutes per job.
ATS Optimization
Increase in estimated ATS match scores due to context-aware semantic tailoring.