Software Engineer at OpenSNZ-Technology
OpenSNZ-Technology · 2025/09 – Present · ongoing · main work
His current role: Software Engineer at OpenSNZ-Technology — designing, building and maintaining production agentic AI systems end-to-end, from rapid prototyping to deployment.
- Ship multi-agent architectures, LLM-driven workflows and agent-to-enterprise integrations to production.
- Architect agent systems around DDD, CQRS and Event-Driven Architecture, with RabbitMQ + Celery as the async backbone — containerized with Docker Compose behind Traefik and Authentik.
- Ground agents in document-heavy enterprise data (LandingAI, Amazon Textract), then deliver them through CopilotKit / AG-UI for human-in-the-loop interaction.
- Orchestrate with LangChain/LangGraph; trace, evaluate and observe agents in production with LangSmith.
- Mentor and supervise interns on architecture decisions, LangGraph/MCP patterns and evaluation practices.
Multi-Agent Systems · LangChain/LangGraph · LangSmith · MCP · CopilotKit (AG-UI) · DDD · CQRS · EDA · RabbitMQ · Celery · Docker · Traefik · Authentik · LandingAI · Textract
AiCA: Pharmaceutical Catalog Platform
OpenSNZ-Technology · Since 2025/02 · released · main work
AiCA, a multi-agent platform for intelligent pharmaceutical catalog analysis — automating data extraction from raw files and coordinating specialized AI agents for querying, cost aggregation, reporting, and decision support. Started during his internship, now in production.
- Started it as an intern in early 2025, when AI agents were still mostly theory; still building it today.
- Designed a modular, scalable, and maintainable agent-based architecture.
- Coordinated specialized agents for extraction, querying, cost aggregation, reporting, and decision support.
- Grounded agents with a knowledge graph and RAG over raw catalog data.
LangChain/LangGraph · LangSmith · Multi-Agent Systems · Knowledge Graph · RAG · MCP · CopilotKit (AG-UI) · Next.js · Python
i18n-ai
codeonym-oss · since 2026/10 · released · personal project
Generates a website's locales from its source locale tree, with LLM agents that keep every placeholder intact. A Python command line with a full-screen form.
- LangGraph classifies every file in parallel, then runs one agent per file and locale, each translating its messages in batches.
- Every translation is checked against rules; a rejected one goes back to the model with what is wrong, until it passes or runs out of attempts.
- Detects the project's i18n library (i18next, next-intl, FormatJS, vue-i18n, Angular, Rails, Django, Flutter and more) and where its locales live.
- Reads and writes JSON, YAML, ARB, gettext PO and XLIFF; works with any provider LangChain's init_chat_model knows.
- Incremental: existing translations that pass the rules are kept; only missing and broken ones are translated.
Python · LangChain/LangGraph · LLM Agents · CLI · TUI · i18n
github.com/codeonym-oss/i18n-ai ↗docs ↗Mediary
codeonym-oss · since 2026/08 · released · personal project
A typed, decorator-driven mediator and CQRS library for Python — handlers, pipelines and notifications discovered by package scan.
- Requests are marked with @request and handlers with @handler; one mediator.scan wires the whole package.
- Typed end to end: sending a request is typed as its declared result.
- Pipeline behaviors wrap handlers; logging, retry and timeout ship ready-made, OpenTelemetry tracing as an extra.
- A CQRS pack (@command, @query, @event), notifications, streams, and dishka and FastAPI integrations.
- Zero dependencies, Python 3.11+, on asyncio or trio; every README example runs in CI.
Python · CQRS · Mediator · asyncio · OpenTelemetry · FastAPI · dishka
github.com/codeonym-oss/mediary ↗docs ↗groundwork
codeonym-oss · since 2026/07 · released · personal project
A plugin for Claude Code and Codex that plans work before building it: it grills an idea into epics, stories and tickets, keeps every decision as a file, and builds each ticket with approval at every step.
- One question at a time, each with a recommendation; every answer is recorded as a decision file next to the code.
- Decisions are inherited: a ticket reads its story's and epic's decisions and never asks them again.
- A ticket's implementation plan is written right before it is built, against the code as it is then.
- Every write outside the planning folder — issues, branches, commits, pull requests — is shown and waits for a yes.
- This portfolio's own plan was made with it.
Claude Code plugin · Agent Skills (SKILLS.md) · Codex · GitHub · GitLab · Jira · Linear
github.com/codeonym-oss/groundwork ↗acli-py
codeonym-oss · since 2026/09 · released · personal project
A friendly command line for Jira Cloud: a Python port of the Jira side of Atlassian's acli, redesigned around short, predictable commands and a dry-run mode for every change.
- A full-screen issue browser (TUI) and a shell that completes commands, issue keys, statuses, people and JQL.
- Smart queries compile short filters to JQL; plain JQL still works.
- Dry run everywhere: reads still happen, each write is printed instead of sent, enforced by the HTTP client.
- Bulk changes by keys, JQL, filter or file, with a per-issue report.
- Script-friendly output (table, JSON, CSV) and several accounts, with API tokens in the system keyring.
Python · CLI · TUI · Jira Cloud API
github.com/codeonym-oss/acli-py ↗jira-time-tracker
codeonym-oss · since 2026/09 · released · personal project
A cross-platform command line (jtt) that answers one question about a Jira Cloud project: how much estimated work came into being over a period, on which issues, and by whom.
- The estimate is whatever field the organisation tracks: story points, an hours or days field, or Jira's own time fields.
- Reports per issue and per person, with an export.
- Published on PyPI; works the same on Linux, macOS and Windows.
- Every command has a documentation page with a recording of its output.
Python · CLI · Jira Cloud API · PyPI
github.com/codeonym-oss/jira-time-tracker ↗docs ↗design-skills
codeonym-oss · since 2026/09 · released · personal project
Design skills for Claude Code with the whole toolchain in a versioned, headless container: Blender, GIMP, Inkscape, Krita, Scribus, darktable, ffmpeg, ImageMagick and a curated font library.
- Eleven skills: graphic design, illustration, photo editing, UI design, typography, print layout, 3D modeling, video editing, screen recording, plus the sandbox and a doctor.
- Each script re-runs itself in the image pinned to the plugin's version, with identical paths, so outputs land next to the user's files.
- Works the same on Linux, macOS and Windows (WSL2) with Docker or Podman — no per-machine setup.
- The 3D assets of this portfolio were built with it.
Claude Code plugin · Agent Skills (SKILLS.md) · Docker · Podman · GHCR · Blender · Shell
github.com/codeonym-oss/design-skills ↗flow-reel
codeonym-oss · since 2026/11 · released · personal project
Turns an end-to-end script into a narrated, captioned demo video: deterministic, in a container, with a local voice. A person or a coding agent writes an ordinary Playwright script plus flow-reel words (comments to speak, sounds to play), and the composer replays it on a virtual display and records the screen. Published on npm.
- A demo is code: one command (npx flow-reel compose) replays the script in Docker or Podman, times every step, voices each comment and fits the video around the speech.
- Polishes the picture with a drawn cursor, click ripples, automatic zoom and interface sounds, and writes a 1080p video, captions, a timed transcript and an event log.
- Deterministic: the same script gives the same result every time, with no retakes and no voice-over drifting out of sync.
- Local and offline: speech comes from a pinned local model, so no API key is needed.
- Agent-friendly: an agent that can write a Playwright test can write a demo.
- Built to extend: a small core with ports for E2E tools, voices and storage.
TypeScript · Node.js · Playwright · Docker · Podman · Speech synthesis
github.com/codeonym-oss/flow-reel ↗GSM Agent Kit
codeonym-org · since 2026/09 · in development · personal project
A modular toolkit for SIM-capable cellular devices and capability-based workflows, with hexagonal architecture. Under development; public source, all rights reserved.
- A Python workspace of 13 packages: domain and capability contracts, application workflows with CQRS, device, AI and storage integrations, a server, a CLI and a browser console.
- Device inspection runs through core contracts, Mediary query handlers and Dishka composition, with a simulated adapter and a ZTE MF833T adapter.
- Incoming SMS ingestion uses a SQLite ledger and durable events; outgoing SMS has durable intent and per-part attempts.
- Replies through OpenRouter use a durable queue and bounded persistent conversations.
- New device adapters inherit a shared contract test suite.
Python · Hexagonal Architecture · CQRS · Mediary · Dishka · SQLite · OpenRouter
github.com/codeonym-org/gsm-agent-kit ↗