Textile CRM
The core CRM for a textile manufacturer, covering orders, models, cutting and sewing flow, and a quality-control module with AQL inspection classes. I work on the whole backend and API layer.
Backend-first software developer, also known as Cipher. I ship B2B platforms, CRMs and factory software for the Uzbek market, and build voice assistants and AI agents on the side. React and TypeScript when the job needs a front end too.
I'm a software developer from Fergana. My day job is B2B software for factories and CRM clients — a textile CRM on Yii2 and PostgreSQL, quality control, CMMS, subcontracting billing and embedded analytics. It's the kind of software where a wrong number on a ledger costs real money, so I care about clean schemas, careful queries and boring reliability.
Outside of work I build the things I wish existed: a voice-first assistant faster than the ones I use, a biometric vault unlocked from my phone, and a Telegram AI secretary that answers for me. Most of it runs on my own VPS, reachable only over a private network.
I started on the front end — React, TypeScript, Ant Design — and moved deeper into backend and infrastructure. That's why I can take a feature from the database to the screen.
Chosen for the job, not the hype. These are the tools I reach for most.
Client and company systems first, then the personal lab.
The core CRM for a textile manufacturer, covering orders, models, cutting and sewing flow, and a quality-control module with AQL inspection classes. I work on the whole backend and API layer.
A Python PDF processor that reads data matrix codes and barcodes, wired into the Yii2 app through a PostgreSQL jobs table and a systemd worker, so heavy work never blocks the web request.
Superset dashboards embedded straight into our React and Yii2 product. The backend issues guest tokens, so users never see Superset or its login.
Maintenance management for a factory floor: machines linked to branches, departments and conveyors, service history, reports with filters, and role-based permissions. Plus a sewing-machine inventory importer.
A Flask and PySerial server that reads weight from industrial scales over RS-232 and exposes it as a REST API, packaged as a Windows EXE. Alongside it, a Kotlin WebView app for company tablets with silent self-update through GitHub Releases.
A self-hosted, voice-first personal assistant that controls my devices from one server and aims to beat Google Assistant on latency. On-device STT and TTS, WebSocket links, and a tiered model setup: a cheap model for everyday commands, a stronger one for hard requests.
A vault for private files on my laptop that only opens when my phone approves it with fingerprint or Face ID. Pairing by QR code, and a signed random challenge on every unlock to prevent replay.
A Telegram AI secretary that answers my messages through Telegram Business. Identity middleware, a persona builder, a fact extractor and owner commands, with switchable Groq and OpenRouter providers and Uzbek-language handling.
A self-hosted AI agent on my own VPS with a custom persona, built on the open-source Hermes Agent framework. Next step: plug it into Milo over MCP so it can handle the requests that need real reasoning.
My personal second brain: one memory store that my agents and tools read from and write to over MCP. It holds context, projects, daily logs and expenses, and my Claude Code sessions are logged into it automatically. Private by design, reachable only over Tailscale, with Attest planned as its biometric lock.
A Python desktop assistant that opens and closes apps, manages files and automates tasks by voice or text. Offline speech recognition, Coqui TTS, fuzzy intent parsing and adaptive learning, tuned to keep latency low.
Responsive admin dashboard for groups, schedules, teachers and students, with CRUD workflows, secure authentication and REST integration. Clean state management with Redux Toolkit and React Query.
A Telegram automation bot with a Telegram Web App login flow and a PowerShell setup script for quick deployment. Built on aiogram and Telethon, with a Flask backend and SQLite storage.
Milo, a voice-first assistant meant to react faster than Google Assistant, Attest, a biometric vault for my own files, and Loom, the memory that ties my agents together.
QC and CMMS features for the textile CRM, conveyor and machine linking, and embedded Superset dashboards.
Robotics, quantum physics and a possible move toward aerospace engineering, while I keep going deeper on software.
French, plus better ways to make AI agents useful: memory, tools and tight latency budgets.
Open to remote roles, contract work and interesting side projects. The fastest way to reach me is Telegram or email.
shukurulloismolilov832@gmail.com