I am a Backend Engineer and AI Systems Developer with a passion for building robust, scalable architectures and autonomous agent ecosystems. My engineering focus lies at the intersection of high-concurrency backend services, LLM & vector search integrations, and resilient automation tools.
Whether designing a multi-agent social network simulator powered by vector databases, developing TLS-fingerprint-resistant automation engines, or building AI-driven job matching engines on top of the German Federal Employment Agency API, I prioritize clean architecture, local-first data privacy, and production-ready reliability.
- 🤖 AI & Autonomous Multi-Agent Systems: Experienced in integrating LLMs (Google Gemini, OpenAI, Groq, Ollama) with persistent semantic memory using Qdrant Vector Database and LangChain.
- ⚡ High-Performance Backend & REST APIs: Architecting asynchronous services and microservices using Python (FastAPI, Django, Aiogram 3.x), Go, and Java.
- 🛡️ API Resilience & Reverse-Engineering: Developing reliable scrapers and automation tools with TLS fingerprint resistance (
curl-cffi), proxy rotation pools, and background task orchestration (APScheduler, Redis). - 🐳 Local-First DevOps & Containerization: End-to-end containerization using Docker & Docker Compose, database optimization (MySQL, PostgreSQL, Redis, Room), and clean multi-stage deployments.
Below are comprehensive technical overviews of my core open-source projects. Each system is architected from scratch with custom algorithms, modern frameworks, and production-grade reliability.
🎯 Jobvis — AI-Driven Job Discovery & Semantic Matching Engine
A production-grade job discovery engine for Jobcenter clients and jobseekers in Germany — it ingests candidate CVs, synthesizes optimal German search queries for the official Bundesagentur für Arbeit API, deduplicates postings, and delivers explainable AI-scored recommendations.
| Core Feature | Technical Implementation & Capabilities |
|---|---|
| 🧠 LLM Query Synthesis & Match Scoring | Uses Groq (gpt-oss-120b) via LangChain to translate multilingual CVs and free-form career goals into German BA search parameters (was, wo, arbeitszeit), scoring matches against skills and CEFR German levels (A1–C2) with explainable rationales. |
| 📄 Multi-Format CV Parser | In-memory parsing of PDF, DOCX & TXT (pypdf, python-docx) with control-character sanitization, size validation, and multi-language skill taxonomy extraction. |
| 🛡️ 3-Tier Deduplication Engine | Normalizes German gender markers ((m/w/d), [gn]) and umlauts, computes canonical hashes, and runs fuzzy similarity checks to discard duplicate listings. |
| ⚡ Async BA Jobsuche Client | Non-blocking REST client with connection pooling, exponential backoff, HTTP 429 rate-limit protection, and automatic multi-page fetching. |
| ⏰ Autonomous Background Matching | Twice-daily APScheduler cron jobs with per-user isolation locks keep candidate feeds fresh without manual intervention. |
| 🔐 Auth, i18n & Production Ops | OAuth 2.0 (Google & GitHub) with signed itsdangerous sessions, zero-reload UI in German, English, Ukrainian & Russian, multi-stage non-root Docker build, MySQL 8.4, and Sentry with PII redaction. |
🌊 BlackWave — AI-Powered Social Network Simulator
A next-generation, local-first AI social network simulator where a single human user interacts with hundreds or millions of autonomous AI bots boasting long-term semantic memory and distinct personalities.
| System Architecture | Key Highlights & Value Proposition |
|---|---|
| • Social Network Backend: Built with Django & MySQL, handling user auth, interactive feeds, dynamic content moderation, and a comprehensive Admin Panel. • AI & Memory Engine: Built with FastAPI & Qdrant Vector Database, orchestrating context-aware bot memory and personality traits. • Multi-LLM Integration: Supports OpenAI, Google Gemini, Ollama, OpenRouter, and self-hosted models for hyper-realistic bot reactions. |
• Autonomous Bot Ecosystem: Bots simulate real social dynamics—including fans, haters, neutral commenters, and humorous critics. • Persistent Semantic Memory: Bots remember past posts and evolve their sentiment toward the user over time using vector embeddings. • Audience Growth & Analytics: Simulates organic follower expansion, viral reach, and realistic engagement curves. • 100% Private & Local-First: Complete containerization via Docker Compose ensuring total data control. |
🔍 Keigen — Intelligent Kleinanzeigen AI Search & Automation Bot
An enterprise-grade Telegram bot & automated search engine for Kleinanzeigen (Germany's largest classifieds platform), featuring GenAI query optimization, semantic intent evaluation, and real-time alerts.
| Core Feature | Technical Implementation & Capabilities |
|---|---|
| 🧠 AI Query Expansion | Uses Google GenAI (Gemini) & LangChain to automatically expand a single user prompt into optimized German keyword variations and synonyms to uncover hidden or non-standard listings. |
| 🛡️ TLS Fingerprint Resistance | Implemented with curl-cffi to precisely emulate native Android client traffic, seamlessly communicating with Kleinanzeigen mobile endpoints without triggering anti-bot blocks. |
| ⏰ Smart Concurrent Parsers | Background asynchronous scheduler (APScheduler 3.x + Redis) running recurring monitors with custom price boundaries, PLZ postal code radius filtering, and AI semantic intent filtering. |
| 🌐 i18n Multi-Language Support | Full internationalization supporting English (en), German (de), Russian (ru), Ukrainian (ukr), and Turkish (tur) with persistent user profiles. |
📡 Telezy — Multi-Source Telegram Message Forwarder & AI Filter
A powerful automated message pipeline that forwards, modifies, and filters Telegram messages between chats using your personal account—no bot channel permissions required.
- Advanced Content Processing: Whitelist/blacklist keyword rules, regex pattern replacements, link extraction, and media size/type filtering.
- AI-Powered Summarization & Translation: Integrates LLM APIs (OpenAI/Claude) to automatically translate foreign streams, tag content, and generate scheduled 24-hour group digests.
- Universal Multi-Platform Dispatch: Leverages Apprise to push processed notifications to Webhooks, Discord, Slack, Email, and SMS.
📱 Todoshka — High-Performance Android Task Manager
A sleek, responsive Android productivity application engineered with native Java, Room Persistence Library, and Google's Material Components.
- Clean Architecture: Built on Android MVVM design patterns with high-performance local SQLite persistence via Room.
- Rich Task Organization: Supports priority tags (Low/Medium/High), dynamic category management, instant full-text filtering, and personalized user profiles.
| Project Name | Stack & Tech | Description & Key Highlights |
|---|---|---|
| 🎨 PixelScale | TypeScript React |
Responsive web application for high-quality pixel art upscaling and real-time image processing. |
| 🎧 Easybook Ecosystem | Java Python Docker |
Comprehensive Android audiobook app combined with microservices for Supabase database migration, custom auth proxying, and Docker deployment. |
| 🖼️ NeuroIMG AI Stack | Python SCSS Nginx |
Full-stack AI image generation platform featuring Python backend workers, SCSS frontend, Nginx reverse proxy, and Telegram bot delivery. |
I am always open to discussing backend architecture, AI/LLM integration strategies, microservices design, and innovative open-source ideas. Feel free to reach out via Telegram or explore my repositories!
