WhatsApp AI recruiting agent for Magnum’s store hiring

Magnum Cash & Carry is Kazakhstan’s largest retail chain. azamat.ai built it an AI agent that gets in touch with everyone who applies for a store job. It talks to candidates via WhatsApp in Russian or Kazakh, asks the screening questions, finds a store with an opening near their home and books a trial shift. Every status goes straight back into Magnum’s HR system.

Reconstructed from one of the agent’s chats. The phone number is masked and pay figures are left out.

The agent went live in September 2026. By the time a recruiter sees a candidate, the candidate already knows their store, the terms and the day of their trial shift.

  • Live since September 2026, in Russian and Kazakh.
  • Offers priority stores first, then others nearby, each with the distance.
  • Writes statuses and comments back into Magnum’s HR system under its own “AI Bot” account.

Retail hiring: more applicants than recruiters can call

Magnum is always hiring for its stores: cashiers, cooks, bakers, sales assistants, warehouse staff, forklift drivers. Every applicant goes through the same first conversation. Are you still looking? How old are you? What’s the pay? Which location is closest? Recruiters have that exchange over and over, and meanwhile applicants lose interest. That repeated first conversation is where AI saves recruiters the most time.

Finding the right store is the hard part. People describe where they live however they like: a microdistrict, a street corner, a university, the number of a store they once worked at. Matching that to a store with the right opening used to take a recruiter, a map and a good deal of time.

  • Applications come in faster than recruiters can call people back.
  • The first conversation follows the same script every time.
  • The right store depends on the address, and no two people type an address the same way.

And every applicant has questions

  • Will the trial shift be paid?
  • Can I work part-time?
  • If my bank accounts are frozen, will you still hire me?
  • That’s too far, anything closer?
  • I’m a student, can I fit it around classes?
  • Do you pay an advance?
  • My little sister is 16, can she work too?
  • Is there a staff bus?
  • Where’s Kemel microdistrict?
  • How soon will I hear back?
  • Is there anything without heavy lifting?
  • Main entrance? Will someone meet me there?
  • Is it the store or the deli?
  • Wat vacansies do you have?
  • Can I work as a loader?
  • What documents do I need?
  • Do you hire people with disabilities?
  • Can I swap shifts with colleagues?
  • I need something near KazGU
  • Can I come on Wednesday?
  • Do you fine staff? What for?

How the AI agent screens a candidate on WhatsApp

This WhatsApp AI agent does what Magnum’s recruiters used to do over the phone, without the queue. It messages first, mentions the job the person applied for, asks the screening questions, explains the general terms, finds a store and books a trial shift. Then it records the outcome in Magnum’s HR system, where a recruiter takes over.

The diagram follows one application.

Magnum’s HR system

An application for a cook job

AI agent

runs the first interview

  • screening rules in code

Candidate’s WhatsApp

  1. Messages first, names the job
  2. Asks the screening questions
  3. Finds a store nearby
  4. Books a trial shift

Status in the HR system

Trial shift booked, a comment or a reason for refusal, under the “AI Bot” account

A recruiter takes over from here

How the agent matches each candidate to the nearest store

Candidates rarely type an address a geocoder can read. Search a map for “Tole bi Saina” and you get nothing; “Almaty, corner of Tole bi and Saina” works straight away. So before looking anything up, the agent adds the city, turns a pair of street names into an intersection and asks when a place could mean two things.

Then it checks which stores are hiring for that job. Stores Magnum has marked as priority come first, nearest first. If there aren’t any, or the candidate turns them down, the agent offers other stores nearby with the same opening.

  1. The candidate writes Saina Shalyapin
  2. The agent adds the city and makes it a corner Almaty, corner of Saina and Shalyapin
  3. Yandex Geocoder returns a point Coordinates of the corner Answers are cached in Redis
  4. First, priority stores Every priority store, nearest first
  5. Then, if none fit Other stores nearby with the same opening
Saina Ave Shalyapin St 1 0.9 km2 1.0 km priority · 1.4 km3 1.3 km5 1.6 km 4 1.5 km Candidate
How the agent lists them: the priority store comes first, even if it’s farther away
  1. priority · 1.4 km
  2. 1 0.9 km
  3. 2 1.0 km
  4. 3 1.3 km
  5. 4 1.5 km
  6. 5 1.6 km
A schematic map: the streets and stores are drawn for the example. The order follows the same rule the agent uses.
The candidate writes “Saina Shalyapin”; the agent replies with stores near the intersection, 0.5 to 3.4 km away The same address in a real chat: stores near the corner, nearest first
Screenshot translated into English. Store names and addresses are hidden.

What the agent handles and what stays with recruiters

We agreed the rules with Magnum’s recruiters before the agent spoke to a single candidate. Where the data is clear, the agent acts. Where it isn’t, it asks. Hiring decisions and the security check stay with people. We draw the same line whenever we set up AI for HR and recruiting for other companies.

  • «That’s too far, anything closer?» — Handles on its own. It offers the next stores by distance without asking for the address again.
  • «Tole bi Saina» — Asks first. It checks whether the candidate means the corner of Tole bi and Saina in Almaty.
  • «My little sister is 16, can she work too?» — Handles on its own. It applies Magnum’s rules for younger applicants, which the code enforces.
  • «Too far to travel, I’ll pass» — Asks first. It asks once what didn’t work. If it’s the distance, it suggests other stores first.
  • «(no reply after the reminder)» — Passes to a recruiter. The candidate becomes a call-back task, and a recruiter phones them.

Handles on its own

Clear rules, clear data

  • Messages first and names the job
  • Applies Magnum’s screening rules
  • Finds nearby stores and how far they are
  • Books the trial shift and updates the status
  • Sends reminders and checks in after the shift

Asks first

Rather than guess

  • The city, if the address doesn’t include one
  • Which corner or district the candidate means
  • Why, once, before taking no for an answer
  • Which day suits them, before booking the shift
  • A screening answer again, if it looks off

Passes to a recruiter

People decide

  • The hiring decision and the security check
  • Candidates who stop replying, as a call-back task
  • Numbers without WhatsApp, with a note in the HR system

How the AI recruiting agent is built

It’s a single LLM agent. It reads free text and decides which tool to call next, and that is the difference between an AI agent and a chatbot. The prompt follows a recruiter’s phone screen stage by stage, and the tools do the actual work: reading the candidate’s profile, finding stores, writing statuses. A FastAPI service receives WhatsApp messages, a scheduler keeps the agent in step with Magnum’s HR system and sends reminders, Postgres holds a copy of the vacancies and store hiring requests along with the agent’s sessions, and every turn is traced in Langfuse.

Rules that must not drift live in code: the screening rules and the order in which stores are offered, priority stores before nearby ones.

WhatsApp Candidate messages arrive by webhook
Taskiq HR system sync and reminders on a schedule
AI agent An LLM with tool calls Tools for the HR system, maps and vacancies
One turn: three tool calls, one reply, one trace.
Magnum’s HR system reads · writes Candidate, vacancy, status, comment
Yandex Geocoder + Redis Address into coordinates, cached
Postgres reads Store hiring requests and vacancies
Langfuse writes A trace of the turn
  1. “Saina Shalyapin”: The candidate sends an address
  2. Open the candidate’s profile: Profile and vacancy from the HR system
  3. Find the address: City added, corner found on the map
  4. Rank the stores: Priority requests first, then by distance
  5. Reply: Stores and distances go back to WhatsApp
  6. Write the status: Later in the chat: status “trial shift”
  7. Trace: Every call in the turn lands in Langfuse
What the agent calls most The longer the bar, the more often the agent reaches for that tool.
  1. Find stores with the opening, priority first
  2. Read the candidate’s profile and vacancy
  3. Look up the vacancy details
  4. Turn an address into coordinates
  5. Switch to the job the candidate wants
  6. Write the status back to the HR system
  7. Explain the general terms
  8. Check where the candidate is in the process

Stack

LLM agent reads the chat and calls the tools
WhatsApp Business messages and approved templates
Python · FastAPI the agent service and its webhooks
Yandex Geocoder turns addresses into coordinates
Redis geocoder cache
Taskiq syncs and reminders on a schedule
PostgreSQL vacancy copy and agent sessions
Langfuse a trace of every turn

What changed for Magnum’s recruiters

  • The first conversation runs without a recruiter

    Recruiters no longer ask every applicant the same questions. The agent asks them, explains the terms and books the trial shift.

  • Everything shows up in the HR system

    Statuses, refusal reasons and comments appear in the system recruiters already use, under the agent’s own account.

  • Quiet candidates don’t slip away

    If someone stops replying, the agent nudges them, then hands them to a recruiter as a call-back task.

“The agent is great at picking stores.”

Magnum HR team

“We tested it. Everything works correctly.”

Magnum HR team, on statuses in the HR system

“The priority ones, it always gets them spot on.”

Magnum recruiter

FAQ

Yes, when the first conversation follows a script, as it does for cashiers, cooks and sales assistants. Magnum’s agent messages everyone who applies, asks the screening questions, explains the terms, matches a store and books a trial shift, all on WhatsApp. Recruiters keep the hiring decision and the security check.

A recruiting chatbot walks candidates through a fixed menu of buttons. An AI agent reads free text, including addresses typed any old way, and calls tools: it looks the candidate up in the HR system, turns the address into coordinates, picks stores and writes the status back. Rules that must never bend are enforced in code.

Start with the first conversation, because it’s almost the same for every applicant. An AI agent can message each one, check them against your screening rules, find the nearest location with an opening and book a trial shift, then log statuses and comments in your HR system. Recruiters pick up candidates who already know their store and their trial day.

Yes. Magnum’s agent replies in the language the candidate writes in, Russian or Kazakh. Plenty of people type Kazakh in Russian letters or mix both languages in one message, and Magnum’s HR team tested those conversations before launch.

Candidates already chat on WhatsApp and reply when it suits them, so nobody gets lost over a missed call. The agent writes first without waiting for a free recruiter, and sends the store list with distances, the documents to bring and the dress code in writing. If someone goes quiet, it sends a reminder and then passes them to a recruiter to call back.

Hiring for lots of stores or branches at once?

We build recruiting agents that work inside your HR system: screening in Russian and Kazakh, matching a location to the candidate’s address, statuses written straight back. We start with one job in one city.

Brief (optional)