How I Built a 6-Agent AI Recruitment System on Qwen Cloud in less than 28 Days

#agents #ai #llm #showdev

The Problem

Recruitment is broken. Recruiters spend 80% of their time on
repetitive task, reading CVs, sending emails, scheduling calls.
I wanted to fix that with AI agents.

What I Built

TryB3-Hire is a multi-agent recruitment platform where six AI agents
handle the entire hiring pipeline autonomously.

When a recruiter pastes a job description, six agents activate:

Intake Agent — parses the JD into structured data using Qwen-Max
Market Agent — researches salary benchmarks and talent supply
Sourcing Agent — scores candidates with full reasoning traces
Screening Agent — has multi-turn conversations with persistent memory
Conflict Agent — mediates when agents disagree
Coordinator — orchestrates everything and escalates to humans

The Tech Stack

Frontend: Next.js 14, Tailwind CSS, Framer Motion
Backend: Node.js, Express, TypeScript
AI: Qwen3.7-Max via Qwen Cloud
Infrastructure: Alibaba Cloud ECS

The Hardest Part

Getting the agent memory to work correctly across sessions.
The screening agent needs to remember what a candidate said
in session 1 when they reply in session 3. We solved this by
generating a one-line memory after each exchange and injecting
it into the next session's system prompt.

What I Learned

Qwen3.7-Max's reasoning capability is impressive for structured
extraction tasks. The reasoning trace it produces helped us
show recruiters exactly why a candidate scored 87/100 not
just the number but the complete chain of thought.