The numbers below are person-hours and money taken off a team. On two of these you don't have to take my word for it: the "Proof" button opens Telegram straight to the client — ask them yourself.
Full-cycle agency automation: marketing, sales, operations
The team came with no spec — just "look at where it hurts". I mapped their processes and closed the whole cycle: from winning clients to the team's daily work. This is the fullest example of how I work: not one automation, but a system of several connected parts.
- Lead gen: Telegram audience parsing + warmed accounts + outreach → 20+ clients; parsing companies across 15+ niches (Trustpilot) + email outreach → ~5 more clients
- Own audience-parsing and outreach software instead of a $300/mo paid analog — that's $3,600 a year and no platform limits; I reused the same software on later projects
- Landing page with a services calculator: the request hits the admin's Telegram instantly, and the calculator takes the "work out the price" step off the manager and filters out non-targets before the call
- An SEO site that brought requests in the first weeks
- Internal 24/7 task manager: tasks moved out of chat into one place — deadline, assignee, "who delivered and who didn't" analytics. Instead of "it got lost somewhere in the thread" — visible status and discipline
- Lead control dashboard: conversion of every outreach campaign is visible — decisions on numbers, not on gut feel
- Telegram business bot: control of working accounts and team notifications about processes
25+ clients$3,600/yr saved15+ niches parsedfull cycle end-to-end
Result: 25+ clients brought in by a lead-gen engine I built from scratch, plus $3,600/year saved on subscriptions. One agency's marketing, sales and operations — on autopilot, with one person accountable for the whole loop.
Publishing reviews on Trustpilot at volume instead of a team of operators
A SERM team needed reviews placed on target projects at volume. Classically that's done by a staff of people working by hand in antidetect browsers (ADS, Dolphin) with proxies — slow, expensive, and dependent on whether someone showed up for their shift.
- Full cycle in ZennoPoster: correct antidetect fingerprint, proxies, automatic captcha solving — from entry to a published review with no human touch
- Behaviour randomization: different entry points, different paths through the site, different pauses and actions — the session doesn't look templated
- Publishing a review with the given text and rating on the target project
- Daily report to a Telegram bot: how many reviews went through in 24 hours — stats at hand, no manual spreadsheets
- 24-hour survival check: the script comes back and verifies whether the review is still alive or moderation removed it. That's quality control, not "fire and forget" — you see real output, not attempts
up to 50 reviews/day~8 person-hours/day≈ 1 full headcount24h survival check
Result: up to 50 reviews a day — that's ~8 person-hours of manual antidetect work removed entirely, a full headcount. The ceiling was held by the platform's risk limit, not by the software: throughput scales with thread count — one thread gives dozens a day, twenty threads give hundreds.
High-volume operations
Mass Cloudflare account registration — 2,000+ per month
A media-buying corporation needed 2,000+ quality Cloudflare accounts a month — purchased ones were cheap but died within days, and the process kept stalling.
- Full registration automation (ZennoPoster): correct browser fingerprint, automatic captcha solving, reliable proxies and quality mailboxes as consumables
- Daily batches on a schedule, with the outcome of every operation logged
2,000+ accounts/molive for weekszero manual work
Result: the need for 2000+ accounts a month covered steadily, in daily batches; accounts live for weeks — against bought ones that died in days. The same pipeline transfers to any resource you burn at volume: ad accounts, domains, emails, payment methods.
Browser automation
Mass review reporting on Google
Same team: reviews had to be reported on Google at volume, every day.
- Same pipeline: proxies, antidetect, full automation — from finding the target review to the submitted report
- Several dozen reports a day with no human involved
~30 reports/day1.5-2.5 hrs/day
Result: ~30 reports a day = 1.5-2.5 person-hours daily that used to go into spinning up profiles, waiting for pages to load and hunting for the right reviews.
Data · infrastructure
Data storage and processing platform — 11M+ contacts
The team had accumulated scattered client databases — sitting in files across folders and in chat threads, i.e. dead weight.
- Deployed a server and a platform holding 11,000,000+ contacts (emails, phones, names)
- Import, export, filtering and search; under the hood AI recognises and classifies contacts, filtering out duplicates and junk
- Integrated a vector database: contacts and bases are searched by meaning, not just by exact field match — the platform finds thematically close bases on its own
- Built in an AI assistant: it suggests which base fits the task, helps assemble the right slice and hands it over ready for export
11M+ contactsa slice in secondssemantic searchAI classification
Result: the slice you need is found and exported in seconds instead of digging through files by hand. Dead databases became a working tool for the team.
AI agents · RAG
First-line AI assistant on the company knowledge base
A closed paid community: people kept asking the same questions, and human support simply could not cover 24/7.
- Answers from internal sources — database, website, articles — plus web search when the base has no answer; remembers the whole conversation
- Multimodal: listens to voice messages and replies by voice, recognises images, analyses documents, prepares summaries
- When it doesn't know, it doesn't invent — it hands over to a human
24/7instant answersfirst line without a human
Result: people paying for access get an answer instantly, including at three in the morning — the bot runs 24/7. The "answer the same thing for the tenth time" routine is off the team; a human steps in only where the bot doesn't know.
AI agents · sales, CRM
AI sales assistant that keeps the CRM up to date
First contact with clients ate the managers' time: half of the conversations never reached a deal, but each one still had to be worked and logged in the CRM.
- Writes to clients, advises from the script and the knowledge base, qualifies interest and creates the CRM record itself
- Trained on six months of the team's real chats — it answers the way the team answers, not "like a chatbot"
- Escalates to a human manager exactly when a human sale is needed
trained on 6 months of chatsqualification on the botCRM fills itself
Result: repetitive questions and qualification sit on the bot, the manager joins an already warm lead. Manual CRM data entry after every conversation disappeared as a separate job.
Monitoring · alerts
Event monitoring with instant notifications
The team had to react to news about tracked projects within hours and be first — too many sources to check by hand.
- Parsing per object from several sources at once: Twitter, Telegram, Discord (specific threads), websites — fires almost instantly
- An event appears → the subscriber gets a Telegram notification right away; alert thresholds are configurable
hours → minutes3+ sources in one feed
Result: going through sources by hand used to eat hours every day — it became one feed with alerts and a reaction time in minutes. It only pings when there is actually something to react to.
Web3 / crypto · 1.5 years — the foundation I came into development from
Browser automation and Web3 analytics: 30+ projects, a fleet of 2,500 accounts
A year and a half in crypto: I was responsible for a fleet of 2,500 accounts, automated every browser action instead of a team of operators, and ran analytical market monitoring. This is where my sense of scale, the anti-detect infrastructure skills and my main professional habit come from — verify the outcome after every action instead of trusting that the script worked. From that foundation I moved into AI automation of business processes.
- Automation for 30+ projects (ZennoPoster): the full cycle from registration to daily routines — interaction with DEX platforms (swaps, cross-chain bridges), prediction markets, trading scenarios, testnet activity
- A fleet of 2,500 accounts: each with its own wallet, browser fingerprint and proxy. Behaviour randomisation — different routes, pauses, amounts and order of actions, so sessions are never templated
- The infrastructure behind it: anti-detect browsers, proxies, captchas, wallet accounting and maintenance, balance and status monitoring — so the fleet stays alive instead of "something broke and nobody noticed"
- Analytics and research: daily monitoring of 100+ sources — blockchains, DeFi protocols, exchanges, infrastructure projects. Digging into teams, funding and mechanics, reacting to events fast
- My first parsers were built right here: pulling funding and activity data from websites and blockchain explorers into one picture instead of manually cycling through dozens of tabs
30+ projects automatedfleet of 2,500 accounts~1,990 accounts on one project$1.3M+ in rewards on that fleet100+ sources monitored
Result: the strongest case of that period — a project where automation carried ~1,990 accounts; total rewards on that fleet exceeded $1.3M. The whole infrastructure underneath — wallets, proxies, maintenance, health control — was mine, both the setup and the daily upkeep.