Product engineer building AI execution systems, marketplace automation, and agentic developer tools.
Kognivu · WBDelegateBot · LinkedIn
AI automation
MCP
agent tools
vertical SaaS
developer infrastructure
marketplace ops
I build product systems where LLMs are useful because the workflow around them is strict: typed tools, approval gates, evals, source-backed memory, operational dashboards, and boring verification.
Most of my strongest work is product-first and not always public on day one. Public repositories show the developer-infrastructure lane; private and launched products show the commercial lane.
|
AI life coach and daily planner for goal execution. Kognivu turns ambitious goals into deterministic roadmaps, daily quests, and accountability loops. The public site is live with structured SEO, sitemap, robots.txt, schema data, and llms.txt; the app is moving toward release. Next.js · AI planning · goal execution · SEO/AEO · product launch |
Released Telegram assistant for Wildberries sellers. Production seller-ops bot for orders, supplies, prices, stocks, documents, subscriptions, and daily reports. Current direction: AI/MCP supervisor with read-only tools first, then preview -> confirm -> execute -> audit for high-impact actions. Python · Telegram · Wildberries API · MCP · seller automation |
|
Private alpha memory and audit layer for corporate AI agents. Connects chats, docs, calls, CRM notes, and project context into reviewed memory. The goal is source-backed answers and actions for teams that need provider control, audit trails, and explicit approval boundaries. TypeScript · Next.js · memory systems · RAG · enterprise AI |
Private incubation: conformance suite for coding-agent terminal runners. Fixture-driven checks for shell timeouts, child process cleanup, background servers, terminal output, cwd drift, environment boundaries, PTY behavior, and adapter reports for agent clients and harnesses. Python · CLI · agent infrastructure · terminal semantics · benchmarks |
Alongside public OSS, I build and operate private vertical products: Wildberries seller automation, tender monitoring bots, finance and crypto operations systems, content repurposing SaaS prototypes, and memory/context infrastructure for AI agents. I make projects public only when the public repo itself is useful, maintained, and honest about its state.
| Project | What it does |
|---|---|
| issue-to-agent | Turns GitHub issues into ready-to-run task packs for Codex, Claude Code, Cursor, and Copilot. |
| mergepack | Turns PR diffs into agent-ready merge packets for maintainers and reviewers. |
| loopback-litmus | Checks browser-to-localhost exposure in local AI agent, MCP, and WebSocket control planes. |
| repo-brief | Generates compact repository briefs for humans and coding agents. |
| ci-fix-brief | Condenses noisy CI logs into repair briefs for coding-agent sessions. |
| action-pin-check | Audits GitHub Actions workflows for mutable or missing action pins. |
| readme-command-check | Checks README shell commands before users copy broken quickstarts. |
I also contribute to established Python tooling. Recent public work includes:
| Project | Contribution surface |
|---|---|
| pytest | Test framework behavior and contributor-facing fixes. |
| coverage.py | Coverage tooling investigation and patches. |
| MkDocs | Documentation tooling improvements. |
| Cleo | Python CLI behavior and tests. |
- AI product systems: agent workflows, tool calling, MCP, memory layers, RAG prototypes, eval-driven loops, and approval-gated actions.
- Vertical SaaS: marketplace automation, finance/ops dashboards, Telegram-first workflows, compliance-aware product surfaces, and founder-led product launches.
- Developer infrastructure: CLIs, GitHub Actions hygiene, repo analysis, CI triage, README validation, shell-runner contracts, and maintainer-friendly automation.
TypeScript, React, Next.js, Python, FastAPI, pytest, PostgreSQL, Prisma, Docker, GitHub Actions, MCP, n8n, OpenRouter/OpenAI-compatible tool calling, and practical LLM workflow design.
- Turn AI products into reliable execution systems, not chat demos.
- Build private product work into public proof where it makes sense.
- Keep OSS focused on real developer/operator pain, with tests and clean docs.