Intelligence,
Engineered.
I'm Яссин, an AI engineer. I build production-grade AI systems — I've created RAG pipelines, LLM applications, and generative AI tools that turn messy real-world problems into software people actually rely on. I work remotely with developers and teams across Austria, Norway, Serbia, and beyond. After years building alongside Russian-speaking teams, they started calling me Yezarov — and it stuck.
I don't just use AI.
I build it.
I take messy real-world problems and turn them into AI systems that actually work in production. Not demos, not prototypes — real tools that people rely on every day.
Projects shipped
Industries
Building AI
Enterprise Knowledge RAG — Austria
Four-month remote build with an Austrian team: a retrieval-augmented knowledge assistant over thousands of internal documents. Hybrid search, a reranking layer for precision, and grounded answers with source citations so staff could verify every response instead of pinging a senior colleague.
AI Product Sprints — Norway
Nine months across two different Norwegian startups — I integrated with each team so fast it felt like I'd always been there. LLM chat assistants, a RAG document-search tool, and generative content pipelines. Fast cycles: build, test, and ship AI features in days, then harden the winners for production.
Legal AI Automation
Built an AI system for a law firm that reads through contracts, finds risks, and flags clauses that need attention. Saved lawyers hours of manual review per case. Uses RAG to pull relevant case law and precedents on the fly.
IKEA Employee Onboarding RAG
Designed a retrieval system that new IKEA employees can ask anything — product specs, store policies, assembly guides. Indexes thousands of internal docs and returns accurate answers in seconds instead of digging through PDFs.
Real Estate Valuation Agent
Multi-agent system that pulls market data, analyzes comparable properties, and generates valuation reports. The agents coordinate — one scrapes listings, one runs the numbers, one writes the final report.
Anomaly Detection Pipeline
End-to-end ML pipeline for detecting anomalies in industrial sensor data. Built custom feature engineering, trained isolation forest and autoencoder models, deployed with real-time inference. Catches equipment failures before they happen.
Multilingual Invoice & Receipt Parser
Built an AI pipeline that extracts structured data from messy invoices, receipts, and handwritten docs in 6+ languages — Arabic, French, German included. Combines OCR with LLM-based entity extraction to pull line items, tax breakdowns, and vendor info from scans that traditional parsers choke on. Processes 500+ documents per hour with 97% field accuracy.
Drone Strike Detection — ML
Contributed to a computer vision project with Serbian developers to detect whether a location has been hit by a drone strike using satellite and aerial imagery. Trained classification models on conflict zone datasets to identify blast patterns, structural damage, and terrain scarring.
We talk
You tell me what you need. I figure out if AI makes sense for it or not. Simple as that.
I look at your data
I go through your docs, your databases, your messy spreadsheets. I need to know what I'm working with before I touch anything.
Architecture
I decide what gets chunked, what gets embedded, which vector DB fits, how the retrieval pipeline flows. Everything has a reason.
Quick prototype
I build something that works in a couple weeks. Not pretty, but it runs on your actual data and you can test it right away.
Break it, fix it
We throw edge cases at it. Weird queries, bad data, stuff users will actually do. I fix whatever fails.
Production ready
I add proper error handling, logging, rate limits. The prototype becomes a real system you can trust with actual users.
Ship it
I deploy, set up monitoring, make sure retrieval accuracy stays high. If something drifts I know about it before you do.
Keep improving
New data comes in, requirements change, models get better. I update the system so it stays sharp. It's not a one-time thing.
Got a project?
Let's talk.
I'm always open to new projects and interesting problems. If you need AI that actually works — hit me up.