Alexey Hunt
Available for hire · Moscow / remote

I build AI systems that work without you

Alexey Hunt. I design and ship products built on neural networks — from a Telegram bot in a couple of days to an autonomous platform that finds clients, keeps track of the money and writes its own code. I work solo or assemble a team to match the load.

HUNT / AGI01
In brief

Not concepts, not Figma mockups. Living systems in production.

Everything below can be opened and touched: sites serve pages, transactions sit in a public explorer, databases are populated, videos exist as files. There is nothing here that couldn't be shown during a fifteen-minute call.

36
projects built and brought to a working state
14
domains in production: sites, platforms, services, bots
0 TRX
sender fees on USDT transfers — gas is paid by an energy sponsor
14,134
leads collected and scored by the model in our own client-hunting system

What I cover

Web platforms and websites · Telegram bots and mini-apps · AI assistants and agent systems · parsers and data collection · CRM and admin panels · desktop apps · infrastructure and deployment.

How I deliver

With sources, deployment and support. Estimate and timeline come in the very first conversation — no managers, no forty-page briefs. The riskiest piece goes first, not the easiest one.

HUNT / AGI02
Web3 · proven on-chain

USDT transfers with a zero gas balance

The user holds only USDT. On TRON a transfer is impossible without TRX, and the market's usual answer is "top up your balance". The working answer is sponsored energy.

Transfer transaction in the Nile explorer

SUCCESSFUL · 29,650 Energy · 345 Bandwidth · 0 TRX burned · nile.tronscan.org

Full run on the Nile testnet: stake → activate → delegate → transfer → reclaim. Every transaction opens in the explorer.
Walkthrough of the demo stand with links to all five transactions →
TronWeb · TRC20 · Energy & Bandwidth delegation · Node.js
HUNT / AGI03
B2B · runs around the clock

A system that hunts clients by itself

A seven-step pipeline: find platforms → collect posts → model screening → database and drafts → sending → reply tracking → failures return to the queue. No purchased lists, no blasting the whole list.

14,134
leads in the database
3,089
messages sent
2,578
model calls for screening
18
people replied personally
Python · SQLite · LLM scoring · Chrome DevTools Protocol · scheduler & watchdog · Telegram
HUNT / AGI04
Products in production · 1 of 2

Not websites — working services

Hunt Motors

Hunt Motors

Turnkey auto boutique: storefront with filters, content and lead admin panel, a Telegram bot for receiving and routing leads, custom design system.

next.js · postgres · bot
Learning platform

Learning platform

22 videos and five-plus hours of material with transcripts, an AI mentor aware of lesson context, full-text search, personal progress and notes.

llm · search · video
Voice rooms

Voice rooms

Rooms joined by a three-digit number, direct peer-to-peer connections, our own TURN server for difficult networks. Audio is never stored anywhere.

webrtc · turn · node
Food analysis by photo

Food analysis by photo

A vision model recognizes the dish and ingredients, calories and macros are computed on top, an LLM builds dietary recommendations. No product handbook involved.

vision · llm
HUNT / AGI05
Products in production · 2 of 2

Tools that grew out of an actual task

KADR Studio

KADR Studio

Raw footage is split into scenes, meaningful fragments selected, an edit plan assembled automatically, renders dispatched to a worker queue.

video · queue
Where I Am

Where I Am

Place suggestions by geolocation and ready-made routes. Works on web and in the bot; a worker keeps the database fresh with no manual care.

geo · worker · bot
Niche research

Niche research

A longread with scroll animation and visualization: where the money is, which pain costs most, where to start. A map instead of motivation.

longread · animation
Visa matcher

Visa matcher

A multi-page site with per-country and per-program breakdowns. Applications reach the owner in Telegram with the client's answers attached — the manager sees full context.

matcher · leads
HUNT / AGI06
Web · industry templates

Landing pages that filter, not just gather traffic

Each one targets a specific industry objection and adapts to a client within days: texts, prices and photos change — the frame stays.

Law firm template

Law firms

Metrics above the fold, practice areas with price ranges, an honest "if we lose" block, success fee billed separately.

Dentistry template

Dentistry

Calm medical palette, price on every service card, FAQ covering pain, cost and installments.

Real estate template

Real estate

Dark premium palette, residence cards with floor area, a transparent escrow deal scheme.

Restaurant template

Restaurant

Menu as text, not a picture. Warm print palette, a section on kitchen and process, one-click booking.

HUNT / AGI07
Media · from raw takes to master

A production line for entire videos

A raw talking-head take becomes a finished vertical video with no manual editing: transcription, speech cutting, subtitles, color, music, intro card, stock inserts matched to each scene's meaning.

Frames from a finished video

Contact sheet of a delivered video: 1080×1920, 47.8 s, eight checkpoints across the timeline

ffmpeg · Whisper · NVENC · stock sourcing via API · batch rendering
HUNT / AGI08
Stack

How I cover a task from start to finish

Languages & environments

Python, JavaScript and Node.js, TypeScript, Kotlin, SQL, Bash and PowerShell. Next.js, Vite, React. Windows, Linux, nginx, systemd.

Web3

TronWeb, TRC20, Energy and Bandwidth delegation, gas sponsoring, node and explorer work, testnets, signature security.

Data

PostgreSQL, SQLite, schemas and migrations, queues and statuses, idempotent retries, crash-safe append-only journals.

AI in production

Model scoring and screening, prompt pipelines, vision, search across materials, MCP servers, agent harnesses, token spend control.

Automation & scraping

Chrome DevTools Protocol, Playwright, Chromium extensions, brittle-markup workarounds, daily platform limits, watchdogs and schedulers.

Media & realtime

ffmpeg at filter and codec level, NVENC, Whisper, speech synthesis, WebRTC with own TURN, batch rendering, automated quality gates.

HUNT / AGI09
How I work

The difference isn't the stack. It's discipline.

I state a machine-checkable result first, then build. Before an expensive step, what exactly will become true gets written down, and verification runs by machine, not by impression. A miss exposes a false belief — so it doesn't repeat.
I hunt causes, not symptoms. An app kept returning "service unavailable in your region". Tunnel and routes were innocent: a separate Go process inside never reads the system proxy map at all. Fixed with one environment variable — unreachable by trial and error.
I never swap work for a stub. No mocks instead of integrations, no "it should work". Whatever isn't done is stated as such, with the reason. Delivery includes sources, deployment and support.
I look with my eyes at everything a human will see. A screen, a frame, a report isn't done until I've watched it as a viewer. Automated checks catch rules, not impressions.
How we start. Fifteen minutes of conversation to understand the task. Then a short spec with price and deadline — you'll see right away whether we match. The riskiest piece goes first: if it doesn't fly, better to learn that in week one.
HUNT / AGI10
Community · Telegram

Night City 👾

A blog I run alongside client work: neural networks, product, and the vibe of Silicon Valley. 422 readers — none bought, all came for specifics. I write only what I deployed and tested on my own projects.

What it covers

Replacing paid AI subscriptions with free and open-source alternatives — including where the swap is honest and where quality drops. Tools and skills for coding agents. Western lead generation. Teardowns of other people's products and of my own mistakes.

What you won't find

Motivation, "success porn" or retold news. Every post is either a working instruction or a measured result. If a tool turned out empty, that gets said plainly too.

Why a client should care: the channel shows how I think and how I verify — before we start working together. There's also a live chat where you can ask anything without starting a private thread.
Channel @aleksei_hunt Channel chat @aleksei_hunt_chat
HUNT / AGI11
Estimates and timelines — in the first conversation

Describe your task in a DM

I answer personally — no managers, no forty-page briefs. A bot, website, CRM, parser, AI assistant, department-wide automation, a full agent system — I take on any complexity and carry it through to results.

Telegram @hunt_tg WhatsApp +7 968 417-24-24 alexfarley555@gmail.com hunt-agi.ru
HUNT / AGI12