- Case study · Bots
- 8 technologies
Meowtrics Team: hiring with AI screening and a Telegram bot
Elvisupdated 6 min read
An HR system for hiring through Telegram: candidates come to the bot, AI reviews their applications, and the team lead sees the hiring funnel and traffic sources.
The challenge
Hiring came in as a steady stream through Telegram. Candidates wrote in DMs, applications got lost in the chats, and nobody could say how many people were at which stage. There was also no way to see at which step people dropped off.
The second issue was job ads. It wasn't clear which channel brought the best candidates and which one just piled up applications that got rejected later.
The recruiter spent time on initial screening that could be handed to AI. New hires had to be trained by hand.
The solution
I built a two-part system in one Turborepo monorepo. The web panel runs on Next.js 15 and TypeScript, and the candidate bot runs on Grammy. Data is stored in PostgreSQL through Prisma.
The bot collects applications and walks each candidate through the stages. OpenAI filters out poor fits, and the panel shows conversion at each step of the funnel. Charts are built with Recharts, and data exports to Google Sheets.
The HR manager edits the bot's texts and the AI prompts in the panel. I did this on purpose so hiring doesn't depend on a developer for every change.
How the system works
Recruiting funnel at a glance
The main screen shows the state of hiring: how many candidates are at each stage, overall conversion and who made it to the end. The data updates on its own, without Excel.
There's separate analytics per recruiter, so the team lead sees how each of them performs. Candidate data exports to Google Sheets, which helps if part of the team is used to spreadsheets.
Telegram bot for job applications and AI candidate screening
The candidate fills out the application right in the bot, and every application goes into the database. OpenAI reviews applications and filters out poor fits, so AI does the initial screening and the recruiter works with those who pass.
The candidate list has filters by stage: who's at the welcome step, who has finished training, who's archived. One click opens the full history of the application and the candidate's actions in the bot.
Source tracking: which channel brings in candidates
Each ad channel gets its own link. The system counts how many people clicked, how many applied and what percentage made it to the final stage.
This shows which channel works better, and the job ad budget can be shifted based on facts. For the team lead it's the same question as in product advertising: how much does one good candidate from each channel cost?
AI training for chatters in the bot
Candidates practice talking to fans right in the Telegram bot. Instead of a real fan, an AI with a set personality and profile replies, and the manager sees every conversation and its score.
The system logs every AI request: who asked, how many tokens it used and how fast the reply came. The logs show where the AI is slow or makes mistakes, and how much it costs.
Bot settings without a developer
In the panel the HR manager changes the bot's texts, sets up the shifts candidates choose from and edits the AI prompts. Changes apply immediately. If the welcome message or an application question needs to change, HR does it without a developer.
Features
- Candidate conversion funnel
- AI application screening (OpenAI)
- Telegram bot for collecting applications
- Analytics by recruiter
- Export to Google Sheets
Screens
SCREEN 1
The hiring funnel at a glance
The main screen shows the full state of hiring: how many candidates are at each stage, the overall conversion and who made it to the end. Nobody keeps Excel sheets: everything updates automatically.
SCREEN 2
Candidates and their applications
A list of all candidates with filters by funnel stage. You see who is at the welcome step, who finished training and who is archived. Click a candidate to open the full history of their application and their actions in the bot.
SCREEN 3
Traffic source trackers
Each ad channel gets a unique link. The system shows how many people clicked through, how many applied and what share made it to the final stage. It is clear which channel works better.
SCREEN 4
AI trains operators
Candidates practice talking to fans right in the Telegram bot, except that instead of a real fan, an AI with a set personality and profile replies. The manager sees every conversation and its score.
SCREEN 5
Bot texts and settings
An HR manager can change the bot's texts, set up shifts for candidates to choose from and edit AI prompts without a programmer. All changes apply immediately.
SCREEN 6
AI logs and stats
The system logs every AI request: who asked, how many tokens were used and how fast it answered. You see where the AI is slow or makes mistakes, and how much it costs.
Results
- Every application goes into the database and doesn't get lost
- You can see which channel brings in candidates
- AI does the initial screening instead of the recruiter
Tech stack
- Next.js 15
- TypeScript
- Grammy
- OpenAI API
- PostgreSQL
- Prisma ORM
- Turborepo
- Google Sheets API
Who it is for
Companies that hire through Telegram all the time and want to manage hiring like a sales funnel.
Project FAQ
Can the system be adapted to how my company hires?
Yes. Funnel stages, bot texts, shifts and AI prompts are changed in the panel, and the system can be extended for a different hiring flow. It's best to start by describing your current funnel.
What exactly does AI do in the hiring process?
AI does three jobs. It reviews applications and filters out poor-fit candidates, plays a fan with a set personality during training in the bot, and scores the conversation. Every AI request is logged with token usage and response time, so the costs are visible.
How long does it take to launch?
It depends on the number of stages, the integrations you need and whether the bot texts are already written. Message me on Telegram at @ElvisOnlyDev, and I'll estimate the timeline for your task.
What is the HR system built with?
The panel runs on Next.js 15 and TypeScript, the bot on Grammy, the database is PostgreSQL through Prisma ORM, and everything sits in a Turborepo monorepo. Screening and training use the OpenAI API, and exports go to Google Sheets.
Ready to automate your business?
A 30-minute call: we go through your processes, pick what to automate first and estimate the timeline.
- CRM
- Traffic
- Bots
- AI
- Analytics
- Integrations
- Websites
Need something similar
- OnlyFans chatter tracking and payrollShifts, sales, KPIs and chatter pay are calculated for you. Any disputed amount opens down to a single sale.
- Custom software for OnlyFans and Fansly agenciesCRMs, bots, chatter extensions and traffic tracking for agencies that have outgrown spreadsheets and per-seat SaaS.
- Bots · 50+ operatorsOnlyServiceA Telegram bot for monitoring chatters, plus a browser extension: shifts are tracked automatically, and reports show where the revenue came from.
- Bots · auto paymentMila VelvetThe model sells access to a private Telegram channel, and the bot handles the subscriptions: it takes the payment, hands out a personal link, reminds about renewal and removes the fan from the channel when the term ends.
- Bots · 6 criteriaFanTrainerBotIs your new chatter losing sales on real fans? Let them practice on the bot first. The customer here isn't real, but the mistakes and the feedback are.
