AgenticHire

Hire 10 AI Engineers next week on board ASAP. Instead of panicking, let's open ASI:One and type one sentence. 5 Fetch.ai agents recruit, rank, and email candidates — gated by a live FET payment.

Comment

Inspiration

It's Monday morning. The CEO storms in:

CEO: "NEW BIG PROJECT NEXT WEEK. WE NEED 10 AI ENGINEERS IN LOS ANGELES ON BOARDING ASAP!!"

Recruiter Team: "WHAT THE HECK?? HOW??"

CEO: "Using Fetch AI — AgenticHire bro. LET ME COOK. 🧑‍🍳"

Hiring the right people is hard. Reaching the right ones fast is even harder. The average time-to-hire for a technical role is 42 days. You post jobs, sift through resumes, write the same cold email 50 times. It's slow, manual, and by the time you find your candidate — your competitor already hired them.

What if we could type one sentence and let AI agents handle everything?

What it does

AgenticHire is a 5-agent autonomous recruiting system built on Fetch.ai's uAgents framework integrated with the Payment Agent Protocol. A CEO types one message on ASI:One. Five agents wake up and go to work:

  1. Orchestrator Agent — receives the CEO's request, immediately requests a 0.1 FET micro-payment (live on the Dorado testnet), verifies it on-chain
  2. Recruiter Agent — uses ASI:One LLM to auto-generate a professional job description
  3. Talent Scout Agent — discovers qualified candidates from the talent network in any role and any location in the United States
  4. Ranker Agent — uses ASI:One to score every candidate against the Job Description (0–100 fit score)
  5. Outreach Agent — drafts a personalized cold email per candidate and sends it via SendGrid

From CEO's sentence → real emails in candidates' inboxes → on-chain TX receipt. Fully autonomous. No human in the loop.

Live demo result: 4 candidates contacted including 2 Fetch.ai Innovation Lab contributors (Kshipra Dhame and Rajashekar Vennavelli) who received real emails live during our demo presentation. Our agents recruited Fetch.ai's own team.

How we built it

  • uAgents framework for all 5 agents — each runs independently on a separate port with mailbox=True for Agentverse discovery
  • AgentChatProtocol + AgentPaymentProtocol on the Orchestrator — this is what makes ASI:One render the native payment card
  • Critical timing trick: RequestPayment must fire BEFORE any LLM call. If there's any delay, ASI:One renders plain text instead of the interactive card. Payment-first is the core design decision.
  • cosmpy for on-chain verification — we verify the TX hash, recipient wallet, and amount directly on the Dorado testnet
  • ASI:One (asi1-mini LLM) for JD generation, candidate scoring, and personalized email drafting
  • SendGrid for real email delivery to all 4 candidates

Challenges we ran into

  • Payment card timing — ASI:One only renders the payment card if RequestPayment arrives within milliseconds. Any async operation before it (like an LLM classification call) breaks the UI. We had to restructure the entire Orchestrator around this constraint.
  • Session management across async agents — tracking which CEO session maps to which pipeline response across 5 agents communicating asynchronously required careful session dict design
  • cosmpy testnet connectivity — the on-chain verifier occasionally fails on network issues, so we built a graceful fallback that trusts ASI:One's CommitPayment for hackathon reliability
  • Python environment conflicts — running on macOS with conda + pyenv created two different Python environments; python and python3 pointed to different interpreters

Accomplishments that we're proud of

  • Native ASI:One payment UI —The payment card, reject/confirm buttons, and wallet balance display are all rendered by ASI:One natively using our AgentPaymentProtocol manifest
  • Full end-to-end pipeline working — CEO message → FET payment on-chain → 5 agents coordinate → real emails delivered → TX receipt shown. No mocking, no hardcoded responses.
  • Live demo with Fetch.ai team — our agents emailed rajashekar.vennavelli@fetch.ai (a Fetch.ai Innovation Lab contributor who was judging at the hackathon) live during the presentation
  • Built in one hackathon SOLO

What we learned

  • How Fetch.ai's AgentPaymentProtocol and AgentChatProtocol work together — and the critical timing constraint that makes them work in ASI:One
  • How to build resilient multi-agent pipelines where each agent communicates asynchronously using typed Pydantic models
  • How cosmpy verifies on-chain transactions on the Dorado testnet
  • The power of building on Agentverse's marketplace — our Orchestrator is discoverable by any ASI:One user with no additional setup

What's next for AgenticHire

  • Calendar agent — after email outreach, automatically schedule interviews based on candidate replies
  • Real candidate discovery — replace the demo pool with live GitHub/LinkedIn API integration for real-time talent sourcing
  • Agentverse marketplace — list AgenticHire as a paid service so any company can invoke it for a FET fee, creating a self-sustaining agent economy
  • Multi-role campaigns — CEO can say "hire 3 ML engineers AND 2 backend devs" and the system runs parallel pipelines simultaneously

Built With

Share this project:

Updates