Welcome. This is a living portfolio — not a record of finished things, but a demonstration of how I solve problems by thinking out loud with the right tools. The process is the point. Three ways in.
Most portfolios show what you've made. This one shows how you think. Each project is a case study in a particular problem-solving approach — not the endpoint, but the process made visible.
"Uzel" (узел) means hub, or node — the point where threads converge, where systems connect. That's the operating principle here: finding the structural relationships between things that look unrelated.
The question underneath every project is the same: how do you make intelligence legible, auditable, and useful.
Domain-agnostic knowledge graph pipeline. Ingests RSS feeds daily,
extracts entities and relationships via LLM with quality gates,
resolves duplicates, and exports interactive Cytoscape.js
visualizations that surface emerging patterns early. Each domain
brings its own entity types, relations, and sources — the framework
handles everything else.
Currently tracking Film (active), AI/ML (paused),
Biosafety (paused), with Semiconductors in design. Runs on a Hetzner VPS with SQLite +
n8n + Docker. Four graph views: Trending, Claims, Mentions,
Dependencies.
Retrieval-augmented generation system built for semantic search over domain-specific corpora. Demonstrates chunk strategy, embedding pipeline, and retrieval quality scoring with human-readable provenance.
Structured evaluation harness for LLM outputs — behavioral consistency, prompt regression testing, quality gates before promotion. Designed for teams moving from vibe-checking to systematic verification.
Observability tooling for agent-based AI systems. Details coming soon.
View project →Multi-agent orchestration built on CrewAI. Details coming soon.
View project →Every project here was built in deep collaboration with AI. It felt right to make that collaboration visible — not hidden in the background, but a named presence with a defined role.
Uzella is the interface between you and the work. She can route you to the right case study, explain a methodology, or walk you through a technical decision in plain language.
She is the thesis made interactive: human-AI systems where the collaboration itself is intentional and legible — not a black box.
Every project follows the same four-phase cycle — Converse, Constrain, Construct, Curate™ — where human judgment and AI capability reinforce each other instead of replacing each other.
The tenth conversation compounds everything that came before. Each pass through the cycle deposits something the next pass builds on.