AgentHire
Five AI agents. One pipeline. Your entire job search — from profile analysis to interview prep — automated by a local LLM swarm.
Architecture
“Job searching is a pipeline problem. So I built a pipeline.”
The Problem
The job search is broken. You write one resume, blast it to 200 companies, and pray. No one analyzes their own strengths objectively. No one researches companies deeply enough. No one tailors their resume for each role. Not because they don't want to — because it takes 40 hours per company to do it right.
The Approach
AgentHire deploys five specialized AI agents orchestrated by LangGraph: a Profile Analyst, Market Researcher, Match Scorer, Resume Tailor, and Interview Coach. Each agent has a focused role and returns structured JSON. The full pipeline runs in 20-25 seconds on a single GPU, powered entirely by local Ollama inference at zero API cost.
Key Insight
The key wasn't making one agent smarter — it was making five agents specialized. A Market Researcher that only finds jobs produces better results than a general-purpose agent asked to do everything. LangGraph coordinates the handoffs, Zustand manages the reactive state, and SSE streams every token to the UI in real-time. The swarm is greater than its parts.
How it works
The architecture behind the system.
5 Specialized Agents
Profile Analyst, Market Researcher, Match Scorer, Resume Tailor, Interview Coach. Each returns structured JSON through MCP servers. Focused expertise, composable pipeline.
LangGraph Orchestration
Directed acyclic graph coordinates agent execution with conditional edges and state management. Pipeline runs end-to-end or agent-by-agent.
100% Local Inference
qwen2.5-coder:14b on RTX 4090. 37-40 tok/s throughput. Full pipeline in 20-25 seconds. Zero cloud API costs — your job data never leaves your machine.
Real-Time SSE Streaming
Every token streams to the UI via Server-Sent Events. Chat system with session persistence, per-message telemetry showing model, duration, tokens, and tok/s.
AI Resume Parsing
Upload a .txt resume and Ollama parses it into structured Profile JSON — skills with categories and levels, experience, education. Dynamic skill extraction.
4D Match Scoring
Skills, experience, education, and culture fit — each scored independently. Not a single compatibility number, but a multi-dimensional breakdown of your fit.
Built with
