Agentic Development Architect — I build multi-agent systems and tooling that make organizations AI-native, from software-development pipelines to process & knowledge automation.
I build enterprise-ready agentic systems — the pipelines, tooling, and infrastructure that turn vague requirements into high-quality, predictable output and help organizations become genuinely AI-native. My work spans two layers:
- 🧩 Agentic software development — disciplined pipelines of specialized AI agents (Analyst → Architect → Planner → Developer → Reviewer → Security) that take requirements all the way to production-grade code.
- 🏢 AI-native operations — agentic tooling for the rest of the organization: RAG knowledge bases,
business-process & document automation (
transcript → process passport → BPMN → DOCX), and consulting / pre-sales workflows.
Across both, I obsess over determinism (scripts and JSON schemas instead of prompt-magic), TDD / VDD (verification-driven development), and top-notch, reusable skills / workflows / hooks for AI-IDEs and agent platforms (Claude Code, Antigravity, Cursor, Gemini CLI, Codex).
- 🔭 Working on: the Multi-Agent Software Development System and the Universal-Skills catalog — plus the tooling that makes organizations AI-native end-to-end
- 🛠️ Building: complex AI automations and agent- & vendor-agnostic systems — agent skills / workflows / hooks, RAG knowledge bases, and end-to-end pipelines (one case among many:
transcript → process passport → BPMN → DOCX) - 🌱 Learning: context engineering, adversarial verification (VDD), and RAG over local knowledge bases
- 💬 Ask me about: agentic development, building AI-native organizations, AI assistants in software engineering, and skills for Claude Code / Antigravity / Cursor
- 📫 Reach me: kuptsov.sergey@gmail.com
The work I'm most proud of — open and ready to explore:
- Agentic Development Framework — a multi-agent development system: vague requirements → quality code through a strict pipeline of agents (Analyst / Architect / Planner / Developer / Reviewer / Security) combining TDD + VDD methodologies. Stack: Python, Markdown, Claude Code / Gemini / Cursor.
- Universal-Skills — a collection of agent-agnostic "meta-skills" (docx/xlsx/pptx/pdf, MCP builder, VDD review, humanizer, summarizing, auto-improving) shipped with a ready-to-use Claude Code marketplace. Stack: Python, Node.js, Bash.
- obsidian-llm-wiki — a multi-vault knowledge base on Karpathy's llm-wiki pattern: Markdown as the source of truth, SQLite (FTS5) as a 100%-rebuildable cache, RAG-with-citations, and
/wiki-*slash commands. Stack: Python, SQLite, Obsidian.
- macOS Utilities — everyday bash scripts: SSD health monitoring (SMART/NVMe, launchd) and recursive
git pull --ff-only. Stack: Bash.
Production systems delivered for clients. Source is private; I'm happy to walk through them on request.
- Process Passport Generator — a high-graded consulting automation framework that turns raw interview transcripts into structured business-process passports (
transcript → Markdown → BPMN → DOCX). - Elma365 Gantt Widget — automation for task planning in Elma365: a high-performance, interactive Gantt-chart widget (hierarchy, drag & drop, dynamic timeline).
- n8n Agent Skills & Conversation Pipeline — enterprise automation orchestrated in n8n:
audio → transcription (Whisper) → meeting protocol (LLM) → DOCX, with a job dashboard. - Demand Generation Automation — automated demand-generation and commercial-proposal pipelines for pre-sales workflows.
These filters fill in automatically as topics are added to public repositories:
- 🤖 AI Agents · ⚡ Claude Code · 🔁 Automation · 📦 All repositories →
⭐️ If any of this is useful to you, a star on the repo is always appreciated.