Build
One agent. Reads the resume, runs the qualifying questions, books the call.
ADK · Agent Studio · Model Garden
BLUEPRINT · STAFFING · RECRUITING
Candidate Intake & Qualification
Qualify inbound candidates against the role spec, schedule the first call, hold the audit trail.
Qualify inbound candidates against the role spec, schedule the first call, hold the audit trail. Named scope, named timeline, named stack — ADK · Memory Bank · Model Armor · 6 weeks.
Design ceiling
Candidates qualified per recruiter per day from 12 to 40+. Time to first call from two days to under an hour.
The agent system is designed to read the inbound resume against the open role spec, run the qualifying questions, book the first call when the candidate clears the bar, and hand the recruiter a one-screen brief on every conversation that lands on the calendar.
The problem
The recruiter is the relationship. The intake is not.
A 40-person recruiting firm carrying a steady inbound stream is paying its recruiters to do the same first pass on every candidate: read the resume, match it against the open role, ask the three qualifying questions, find a slot on the calendar. The work is repetitive, the bar is codified in the role spec, and the calendar coordination is the slowest part of the day. Cycle time stretches into days and the strongest candidates go cold while a recruiter is still chasing a reply.
The job is not to replace the recruiter. The job is to walk the recruiter into the first call already warm, with the resume read, the qualifying questions answered, and the slot already on the calendar. One agent, role spec held in Memory Bank across the conversation, Model Armor at the gateway on every PII handling step, every booked call linked back to the recruiter who owns the search.
Agent architecture
Four pillars, this blueprint’s stack.
The platform’s four pillars, mapped to the components this agent system actually exercises.
Scale
Role spec and candidate history live in Memory Bank across the qualifying conversation.
Agent Runtime · Memory Bank
Govern
Every PII access policy-checked. Every booked call tied to the recruiter who owns the search.
Agent Registry · Model Armor · Gateway · Identity
Optimize
Eval set walks the qualifying-bar edges and the historical recruiter overrides.
Evals · Observability · Agent Analytics
Engagement · 6 weeks
Week by week.
Fixed scope, fixed price, fixed timeline. Here is what happens when.
Week 1
Discovery and role-spec inventory.
Walk the current intake workflow with the head of recruiting and two senior recruiters. Inventory the open role specs, the qualifying questions per role family, the ATS connector, the calendar feed. Shape the eval set against recruiter accept rate on booked calls.
Week 2
Resume reader and role match.
Stand up the ADK agent that reads the inbound resume, matches it against the active role spec, and surfaces the gaps the qualifying conversation needs to close. First eval pass on a sampled month of historical applicants.
Week 3
Qualifying conversation and scheduler.
Layer the qualifying-conversation flow on top of the reader. Wire the scheduler tool into the recruiter calendar. Every booked call carries the resume read, the question answers, and the recruiter brief.
Week 4
Governance and registry.
Configure Model Armor policies for PII handling on resume content and contact data. Enrol the agent in Agent Registry against the recruiter pool. Stand up the audit trail so every qualified candidate links to the recruiter who owns the search.
Week 5
Staging shadow run.
Run the agent in shadow against live inbound for a week. Compare the qualified shortlist to the recruiter screen with the head of recruiting. Tune the qualifying bar and the eval set against any drift.
Week 6
Production cutover and handoff.
Deploy to Agent Runtime with the recruiter brief landing on the calendar invite. Walk the runbook with the head of recruiting. Four-week post-launch support window for drift watch and recruiter feedback.
What it looks like in code
ADK Python — the agent definition.
The actual shape of the code your team owns at engagement end. Real ADK, real tools, real instruction copy.
agents/intake/agent.py
python
from google.adk.agents import LlmAgentfrom google.adk.tools import FunctionToolfrom .tools import ( fetch_role_spec, parse_candidate_resume, run_qualifying_conversation, book_recruiter_call,)intake = LlmAgent( name="candidate_intake", model="gemini-2.0-pro", instruction=( "You qualify inbound candidates against the active role spec. " "Read the resume, run the qualifying questions, and when the " "candidate clears the bar, book the first call on the recruiter " "calendar with a one-screen brief attached. Never schedule a " "call without the recruiter who owns the search on the invite." ), tools=[ FunctionTool(fetch_role_spec), FunctionTool(parse_candidate_resume), FunctionTool(run_qualifying_conversation), FunctionTool(book_recruiter_call), ],)What you walk away with
Code your team owns. Governance baked in.
Every blueprint hands the engineering team a deployed agent and the artefacts to run it themselves. No black box, no lock-in.
- One ADK intake agent deployed on Agent Runtime, in your Google Cloud project.
- The Git repo your team owns, with the agent code, tools, prompts, and tests.
- CI/CD pipeline wired to your repo, deploying on merge.
- Eval harness with a sampled-month corpus across the qualifying-bar edges and recruiter overrides.
- Governance bundle: Agent Registry entries linking each booked call to the named recruiter, Model Armor PII policies, Gateway access controls.
- Agent Analytics dashboard plus an on-call runbook for the head of recruiting.
Book this blueprint as a pilot.
Two weeks. Named scope. Working agent on Agent Runtime at the end.
Code
Lives in your Git org, owned from commit one.
Governance
Model Armor and Agent Registry on day one.
Speed
Two weeks to a runnable pilot. Eight to production.
Not ready to talk? Take the 4-min readiness assessment