AI engineer & technical SEO · Vilnius, LT
Same engineer, two applications of the same skill: I ship ML and LLM systems to production, and I grow organic search through technical SEO. Python and SQL across the stack, deployed and monitored, not left in a notebook.
role = build → deploy → monitor
domains = [ "AI/ML", "LLM agents", "technical SEO" ]
// two tracks
Supervised and unsupervised modelling, RAG pipelines, and agentic workflows, taken from data to a deployed endpoint with monitoring. Statistics done properly: significance tests, calibration, leakage-safe validation.
Crawlability, site architecture, migrations with redirect continuity, and content systems built to rank. I read the codebase and the analytics, so recommendations ship instead of sitting in a deck.
// selected work
live endpoint // serverless, zero-dependency
Retrieval-augmented QA API on AWS Lambda + API Gateway, with the embedding index in S3. Live HTTP endpoint; answers stay grounded in source documents and decline what the data does not cover. Zero-dependency Lambda, deployed from one script.
Python · AWS Lambda · API Gateway · S3 · embeddings
1,133 chunks // 102 pages, paying client
A retrieval chatbot running for a paying B2B client, grounded in 102 scraped pages and 1,133 knowledge chunks. Chat, lead-capture, and health APIs as microservices, with live monitoring of retrieval quality, latency, and conversion.
Gemini 2.5 Flash · Node.js · RAG · monitoring
human-in-the-loop // tool chaining + gating
An agentic system on LangGraph that chains tools, routes conditionally, and pauses for human approval before acting — the pattern real workflows need, where an agent escalates rather than acts alone. Benchmark scenarios included.
LangGraph · LangChain · ChromaDB · FastAPI
+5.9 Dice, p<1e-8 // external cohort
Patch-based semi-supervised segmentation of pancreatic cancer on CT (281 patients). The semi-supervised model beat the supervised baseline by 5.9 Dice points on a fully external cohort (paired Wilcoxon p < 1e-8). Oral talk at eStream 2026.
PyTorch · TensorFlow · HPC / SLURM · statistics
95+ Lighthouse // new domain to visibility
Built a new domain to organic visibility: geo-targeted service pages, structured data, and Core Web Vitals work (95+ Lighthouse). Enquiries arrive through lead forms I built and track, so the channel is measured in leads, not rankings.
Technical SEO · structured data · CWV · lead capture
27,879 impressions // one article, 3 months
Search performance for a technical B2B site in industrial automation and SCADA cybersecurity. A SCADA-attack article earned 27,879 impressions in three months against established publishers, and content here earns Google AI Overview citations.
Technical content · GEO/AEO · B2B · Search Console
// stack
Python, SQL, NumPy, Pandas, scikit-learn, PyTorch, TensorFlow/Keras; supervised & unsupervised modelling, statistics (Wilcoxon, bootstrap CIs, calibration).
RAG pipelines, embeddings, prompt engineering, LangChain, LangGraph multi-agent orchestration, human-in-the-loop; Gemini, GPT, Claude, Mistral.
AWS (Lambda, API Gateway, S3), Docker, FastAPI, REST & batch serving, Git/GitHub, GitHub Actions CI/CD, production monitoring.
Technical audits, migrations, site architecture, Core Web Vitals, structured data, GEO/AEO; GA4, Search Console, Ahrefs, Screaming Frog; WordPress, Elementor, Shopify.
// proof
My background started in web development and grew two ways: into applied AI through an MSc and real deployed systems, and into technical SEO through years of client work. The common thread is engineering judgement — diagnose the real problem, respect the constraint, and ship something that holds up in production.
// testimonials
// contact
AI/ML engineering, LLM & agent systems, technical SEO, and production-minded web work. Based in Vilnius, eligible to work in Lithuania.