Operator profile · loaded

I build & operate
production systems
end‑to‑end.

Solo engineer, shipping production code since 2003. I write the code, run the boxes, and pick up the pager. 40+ servers under management across 8+ datacenters, eight production ML services, multi-source data warehouses — kept alive by one person who knows every layer.

NDA always respected · client details never shared
~ / dev / whoami
↓ scroll to interact

Live numbers, not slideware.

Every figure below comes from infrastructure I personally provision, patch, monitor and back up. These aren't claims — they're the contents of a running operations log.

[ infra ]
0+
production servers
8+ datacenters, 3 continents
[ data ]
0B+
analytical rows
2 ClickHouse warehouses
[ ml ]
0
ML services in prod
YOLO · TrOCR · CatBoost · Paddle
[ media ]
0M+
images on own CDN
geo-routed, 4-tier fallback
[ backup ]
0 TB
backups
37 sources, 7-day rolling
[ probes ]
0+
monitored services
5-min probes · own stack
[ ingest ]
0+
parsers in production
multi-source · 24/7
[ tests ]
0+
unit tests on hot path
pipeline coverage

Real infrastructure, mapped.

Stylised view of a single platform I run end-to-end: parsers feed a master DB, ML enriches lots, an internal API serves three consumer products, edge proxies bypass regional blocks, and a backup orchestrator pulls everything home every 30 minutes.

INGEST CORE ENRICH SERVE parser · α high-volume · py parser · β regional · proxies parser · γ node · 9 svc stream · WS real-time · async scheduler 60+ cron jobs MASTER RDBMS · replicated 100s GB · WAL photo CDN 328M+ files · multi-region backup-orchestrator · 60 TB ml · vision YOLO · TrOCR · Paddle ml · predict CatBoost · forecast clickhouse · 2x 4.6B+ rows · events analytics · LLM NL → SQL → chart web · A web · B web · C mobile · iOS/Android via edge · auto-TLS · failover
Ingest (parsers · scheduler)
Core / ML enrich
Analytics / serve
Edge · CDN · consumers
Backup pull (every 30min)

What I actually ship.

Five active verticals. Some are public products with my name on them; others are operational platforms for businesses across LatAm, Asia and Europe. All shipped, all running, all watched by me.

01 parsing

High-volume parsing platforms

End-to-end ingest: 24+ parsers across multiple regions and verticals, normalised into a unified data model, served via internal REST APIs to multiple consumer products and mobile apps.

  • Multi-source feeds normalised
  • ~1000 unit tests on hot path
  • RabbitMQ fanout for downstream
  • iOS / Android via React Native
02 ml in prod

Production ML pipelines

Eight production ML services on a single GPU server. Not Jupyter demos — FastAPI endpoints behind health probes, deployed via Gitea Actions, tracked in MLflow.

  • YOLOv8 car & layout detection
  • TrOCR + YOLO multi-task OCR
  • CatBoost price forecasting
  • PaddleOCR · LibreTranslate JA/KO/ZH
03 data & etl

Warehousing & LLM analytics

Multi-source ETL into 4.6 B+ rows across two ClickHouse warehouses. Custom analytics UI with a natural-language → SQL → chart pipeline powered by GPT and a read-only SQL validator.

  • Multi-CRM normalisation (MT4 · HubSpot)
  • 120-node n8n GPT workflow
  • Real-time OpenAI spend tracking
  • LTV / retention / cohort cubes
04 infra ops

Infrastructure as a discipline

Operations as a first-class deliverable. Two independent monitoring stacks, custom backup orchestration, edge proxies, ISP-grade subscriber control planes (RADIUS, Juniper provisioning), bare-metal hosting because cloud bills are not a feature.

  • FastAPI backup orchestrator · 60 TB
  • 100+ probes across 43 hosts
  • RADIUS / Juniper / BAS-IP control planes
  • Caddy reverse proxies under hard nets
05 real-time

Real-time & fintech

WebSocket pipes that survive proxy bans and exchange rate-limits. Multi-exchange aggregation, copy-trading routing, payment gateways, on-chain wallet monitoring — for platforms where milliseconds and uptime equal money.

  • Multi-exchange WS aggregators
  • Pooled-proxy failover under bans
  • Crypto wallet tracking · alerts
  • Copy-trading routing bots
06 platforms

Ops portals & back-office

Internal workbenches with role-based access, KPI dashboards, approval workflows. Multi-brand mail infrastructure (IMAP/SMTP daemons), recruitment data scrapers, trader desks — boring, lean, kept simple.

  • RBAC ops dashboards · multi-role
  • Mail infra · IMAP/SMTP daemons
  • Recruitment / HR data scrapers
  • Token-based auth · ClickHouse logs

Tool to job, not job to tool.

No religious wars. Node for I/O, Python for ETL and ML, Go where latency matters, PHP for classic sites and admin panels, bare-metal where the cloud bill stops being a feature.

Languages
Python TypeScript Go PHP Bash SQL
Frontend
React React Native Capacitor Vite Tailwind Bootstrap vanilla JS
Backend
FastAPI Flask Express Node PHP (custom + Symfony-ish)
Data
MySQL ClickHouse PostgreSQL Redis RabbitMQ SQLite
ML / AI
YOLOv8 TrOCR CatBoost PaddleOCR PyTorch LibreTranslate OpenAI GPT MLflow n8n Ollama
DevOps
Docker nginx Caddy Gitea GitLab CI rsync cron systemd

What I can own for you.

Roles that map to deliverables. I can step into any of these on day one and stay there for years.

System architect

Design data flow, service boundaries, ownership lines, failure modes. Pick the right database and the right protocol the first time, not the third.

Full-stack engineer

Write the React component, the API endpoint, the SQL migration and the deploy pipeline in one branch. No handoffs, no waiting on another team.

Infra operator

Provision bare-metal, set up TLS, harden SSH, schedule backups, watch metrics, get paged at 3 AM and fix it before sunrise. Day-zero through year-five.

Data engineer

Build ETL that survives schema drift. Tune MySQL replication, partition ClickHouse, model warehouses for analysts that actually want to query them.

ML in production

Not "we'll train a model." Models behind FastAPI, version-pinned, health-probed, retraining tracked, costs monitored. The model is the easy part; shipping it is the job.

Solo founder bridge

For founders without a CTO: I'll spin up the whole technical side, take it to product-market fit, and either hand it cleanly to a team or keep running it.

Cross-border by default.

Servers across 3 continents, products in 7+ countries, fluent in shipping under sanctions and through every kind of regional friction.

🇯🇵
Japan
auctions · translations
🇲🇽
Mexico
fintech · ops portals
🇮🇱
Israel
domain platforms
🇰🇷
Korea
marketplace ingest
🇨🇳
China
marketplace ingest
🇺🇸
USA
cold storage · CDN
🇺🇦
Ukraine
telecom · billing
🇵🇱
Poland
B2B platforms

How I work.

Not a manifesto. The actual heuristics that let one person hold this much production code in their head.

P · 01
Build → operate, not build → handoff.

I write code knowing I'll be the one paged a year later. That changes how careful you are about logs, defaults, and silent failures.

P · 02
Every dependency is a debt.

Vanilla PHP, FastAPI, nginx will outlive any npm apocalypse. I add libraries when they pay rent, not when they're popular.

P · 03
Backups before tests.

Tests fail — you fix the code. Data lost — you're done. Backup orchestration is the first thing I set up, not the last.

P · 04
If the alert is silent, the system is healthy.

I don't watch dashboards. Systems write me on Telegram when something breaks. Anything else is dashboard theatre.

P · 05
The job is not the model — the job is shipping it.

Anyone can train a YOLO. Putting it behind FastAPI, version-pinning, monitoring drift, autoscaling on a GPU box you actually own — that's the job.

P · 06
Cheap, boring, owned.

Bare-metal that I rack-monitor myself beats a cloud bill that doubles every quarter. Boring tech that I've run for a decade beats hype tech the conference loved last week.

P · 07
What I don't take on.

Legacy CMS rewrites (WordPress · Bitrix · Joomla). Pure design without code. Native Swift / Kotlin from scratch. Gambling, casino, NFT / Web3 launches, equity-only deals. Micro-tasks. Cowboy-rescue jobs without discovery — we look at the state first, then talk numbers.

P · 08
What I take strictly.

Long-running projects where I own code AND operations. Serious platforms under NDA. Systems built to run for years. With a normal contract and a real client behind it.

Working on something serious?

I take on a small number of projects at a time. If you need someone who can own the full stack from the bare metal up — write.

guest@devby.pro:~$ contact --priority high --about "your project"
— or via telegram —