βββ ββββββββββ ββββββ βββ ββββββββββββββ βββ
βββ ββββββββββββββββββββββ ββββββββββββββ βββ
βββ ββ ββββββββββββββββββββββ βββββββββββββββββββ
βββββββββββββββββ βββββββββββ βββββββββββββββββββ
βββββββββββββ βββ βββββββββββββββββββββββββ βββ
ββββββββ βββ βββ βββββββββββββββββββββββββ βββ
π οΈ AI Infrastructure Engineer Β· π Building from Kazakhstan Β· π€ Multi-agent systems
I build infrastructure for AI systems β not the demos, the machinery underneath. Multi-agent orchestration, streaming primitives, IoT platforms, animation components. Everything is open source.
Working on problems that matter to Central Asian developers: address data nobody cleaned up, AI tooling never localized for non-English devs, infrastructure the ecosystem was missing.
Currently: building agent-kit β a TypeScript framework for multi-agent Claude systems.
| Repo | What it solves | Stack | |
|---|---|---|---|
| π§ | rag-service | RAG API β chunking, embeddings, vector search, source-cited answers + hallucination guardrail | Python Β· FastAPI Β· Docker |
| π§© | agent-flow | Async multi-agent orchestration β planner β parallel workers β synthesizer | Python Β· FastAPI Β· AsyncIO |
| β | agent-kit | Multi-agent orchestration β pipeline, parallel, supervisor, swarm flows | TS Β· Claude Β· Anthropic SDK |
| π€ | ai-saas-starter | Claude streaming + auth + billing, Edge runtime, deploy in minutes | Next.js Β· tRPC Β· Prisma |
| β‘ | react-ai-hooks | useChat useStream useCompletion β streaming primitives for React |
React Β· TypeScript |
| π± | terra-platform | Agricultural IoT β SSE sensor feeds, real-time threshold alerts | Next.js Β· SSE Β· EventEmitter |
| π± | telegram-ai-bot | Telegram Γ Claude with Redis conversation memory + streaming | Telegraf Β· Redis Β· Anthropic |
| π¨ | motion-primitives | Framer Motion: magnetic buttons, scroll-parallax, text reveal | React Β· Framer Motion |
| ποΈ | nextjs-dashboard | Production admin template β tRPC v11 + next-auth v5 + Prisma | Next.js Β· PostgreSQL Β· Redis |
| π°πΏ | kz-addresses | Kazakhstan address data β trilingual KAZ/RUS/ENG, all 20 regions | TypeScript Β· JSON Β· GIS |
Principle 1 β No magic. Internals should be readable, forkable, understandable.
Principle 2 β Real problems first. If I can't deploy it myself, it doesn't ship.
Principle 3 β Open by default. 1,000 forks > 1 paying customer.
Languages TypeScript Β· Rust (learning) Β· Python
Runtime Node.js Β· Edge (Cloudflare Workers / Vercel)
AI Claude API Β· multi-agent Β· streaming Β· tool use Β· RAG
Frontend Next.js 14 Β· React 18 Β· Tailwind Β· Framer Motion
Backend tRPC v11 Β· Prisma Β· PostgreSQL Β· Redis
Infra Docker Β· GitHub Actions Β· Railway Β· Vercel
From Almaty, Kazakhstan π°πΏ β one of the most underrepresented tech ecosystems on GitHub.
Most of what I build solves problems I couldn't find good open-source solutions for in Central Asia.
The intersection of AI + local context is where interesting problems live.