AI agents & LLM systems
Agents that take real actions, multi-agent pipelines, retrieval, and evaluation on every endpoint.
- OpenAI
- Anthropic
- Gemini
- Tool use / MCP
- Multi-agent
- RAG
- Embeddings
- LangChain
- Langfuse
Python Backend & AI-Systems Engineer · 3+ years
I build backends that hold up, and AI agents that act reliably.
I'm a backend-focused engineer in Germany. I ship event-driven FastAPI microservices, Kafka and Redis pipelines, payment and messaging integrations, and the infrastructure that lets AI agents take real actions safely.
01
About
I've worked both ways: as part of a team building a microservice platform together, and as the sole backend developer on projects, responsible for everything from architecture and the data model to deployment and support.
For the last three years I've built backends for fintech and SaaS products: bank and payment integrations, event-driven microservices on FastAPI and Kafka, and multichannel messaging. Much of that work has been designing the architecture of AI-powered systems that have to be GDPR compliant, low-latency, secure and reliable in production.
As a side project, I'm building repeatandlearn.com, an enterprise-level spaced-repetition SaaS on Telegram, Discord and a PWA. Owning it end to end, from the delivery dispatcher to pricing, ads and support, taught me how important it is to listen to users, gave me the right perspective on what to focus on when building features, and taught me many other lessons I bring to every project.
- Role
- Backend / AI engineer
- Experience
- 3+ years
- Core stack
- Python · FastAPI
- Focus
- Event-driven · AI agents
- Location
- Germany · UTC+1
- Work rights
- Authorized in Germany
- Format
- Remote · Hybrid · On-site
- Languages
- EN C1 · RU · UK · DE B1
02
What I work with
Backend & event-driven
Async Python services and microservices that stay predictable under load.
- Python
- asyncio
- FastAPI
- Pydantic
- Kafka
- Redis Streams
- Taskiq
- Celery
- WebSockets
- pytest
Data & storage
Relational models, migrations, idempotent transaction models and data collection.
- PostgreSQL
- SQLAlchemy 2.0
- Alembic
- Supabase
- MongoDB
- Neo4j
- SQLite
- Web scraping
Voice AI
Speech-to-text and text-to-speech for real-time calls, self-hosted on GPUs or via vendors.
- faster-whisper
- Silero VAD
- Deepgram
- Speechmatics
- ElevenLabs
- RunPod
Cloud & DevOps
Shipping and running services with containers, CI/CD and monitoring.
- AWS
- Azure
- Cloudflare
- Docker
- Kubernetes
- Terraform
- GitHub Actions
- Fly.io
- Sentry
- Loki
Frontend
Full-stack when the product needs it, from offline-first PWAs to admin panels.
- TypeScript
- React 19
- Vue 3
- Inertia
- PHP / Laravel
03
Experience
Python Software Developer
Backend and AI infrastructure for AI products in sales automation and accounting.
- Owned the in-app AI assistant of the accounting product end to end: architecture and Python code of a Kafka-based orchestrator running a Codex agent with a 4-agent LLM fallback, with core services exposed as MCP tools.
- Moved speech-to-text for the AI managers' phone calls from vendor models to self-hosted Whisper (RunPod) for GDPR compliance, and built GDPR safeguards for LLM traffic: reversible PII anonymization (Presidio), masked logs and an AI audit log.
- Unified 5 bank, payment and fiscal providers into one idempotent transaction model and built its classification pipeline (rules, embeddings and an LLM), which sorts 99% of transactions without manual review.
- Built the SMS / iMessage channel (Twilio, LoopMessage), later sold as a $100/month add-on, and shipped email (Gmail API, Microsoft Graph) with an AI spam filter. Multichannel outreach lifted conversion by 22%.
- Built 24/7 multichannel alerting on Loki and Sentry (Telegram, Slack), plus an external Cloudflare Worker that checks the services from outside, so outages get caught even if our own monitoring goes down.
- Owned optimization of 16 of 30 microservices: faster, safer inter-service calls and slow work in background jobs.
Junior Python Developer
AI-driven B2B debt collection.
- Built an AI lead-generation tool that collects speakers from US event platforms with no official API and filters them by ICP, surfacing hundreds of leads per month.
- Built an LLM-powered Excel repair tool that maps arbitrary client debt files to our template, cutting onboarding from ~2 hours to under 5 minutes, letting ops onboard 3x more portfolios.
- Built a RAG dispute autoresponder for debtor emails, evaluated with LLM-as-a-judge (Langfuse).
04
My own project
Live · own SaaS
Jul 2026 - present
RepeatAndLearn
A spaced-repetition SaaS that delivers knowledge cards to Telegram, Discord and its own offline-first PWA. Users answer "Remember / Forgot" right in the chat, and the algorithm moves each card along a ladder of intervals. AI assistants condense texts, answer questions about cards, search the web and translate cards for language learners.
I built and run it solo: backend, frontend, billing (Paddle), analytics, support and GDPR compliance, and grew it through Meta ads and organic content.
- Designed a horizontally scalable delivery dispatcher across 25 bots (20 Telegram, 5 Discord).
- Built an automated pipeline that generates dozens of Instagram posts a day across 5 Instagram channels.
- FastAPI
- PostgreSQL
- Supabase
- React 19
- TypeScript
- PWA
- Paddle
- Fly.io
05
Education & languages
BSc, Computer Science and Intelligent Systems
National Technical University "Kharkiv Polytechnic Institute"
- English C1
- Russian Native
- Ukrainian Native
- German B1
Contact
Hiring? Let's talk.
I'm looking for a full-time backend or AI-engineering role in Germany or remote across Europe. Send a note and I'll reply within a day or two.