Styloma
Sure, let me tell you about it…

What if you could search for clothes by showing what you mean, not by trivial filtering.

Lead solo developer

Search fashion, visually

You want a dress just like that one; you visit Zara, filter by category and size. Then H&M, same filters. Then ASOS. Each site separately…

Visual semantic search

I built a dedicated search for exploring many stores at once - allowing you to find what you'd never discover by browsing page by page.

Image search

Upload a photo, iteratively explore similar items, until you find exactly what you're looking for!

Text search

With natural language semantic understanding, you can give even more details.

URL search

Paste product link → find it elsewhere, maybe even cheaper?

Visual searchE-commerceObject embeddingObject detectionReactTypeScriptViteMUI

Event-driven microservices

I architected this as 15+ microservices on self hosted server. Services coordinate through Kafka events - when a new item gets ingested, it triggers embedding generation, duplicate detection, and search indexing automatically.

Data ingestion challenges

Retailers can expose different data formats - some provide APIs, feeds, others require scraping. I constructed a set of configurable parsers handling each source's specific access method and limitations.

MicroservicesEvent-driven architectureRESTJWTWeb scrapingKafkaKubernetesTerraformNixOSLinux serverGitHub pipelinesPuppeteerPostgreSQLRedisClickHouseHelmPrometheusGrafanaSentryExpressTypeORM

Vector embeddings for similarity

A picture is worth a thousand words? I think so, but a pixel match is not enough. We need to go further to understand for example the difference between a wrap dress and a shift dress, not just "blue dress". Since the beginning I knew that embeddings need to capture semantic meaning, not keywords.

Embedding types

Image embeddings

Visual similarity - Individual photos encoded as a vector - "find pictures that match this".

Item embeddings

Semantic similarity - Items encoded as vectors of their properties - "find items that fit this description".

User embeddings

User interest - Past interest encoded as a vector - "prioritize items that I will like".

Hybrid search

And now combine them all to get the individualized exact matches.

Machine learningObject embeddingPyTorchPythonONNXOllamaRunPod

Human (in the) loop

The agentic development sub-system

Agentic developmentToolingMCPValidationVerificationQACI/CDDockerGitHub ActionsVitestPlaywright

Bibliography

2025JulSep2026AprJun2027;Design;Start work;Initial testing;Finalization;Beta Launch
  1. Gmiterko, D., “Styloma β”, 1.0.0, 2026, source is not open, but you can test it yourself at styloma.deegmiterko.com ▷
  2. Gmiterko, D., “Eventually consistent cyclical agentic-aided workflows”, Brno CZ, ask me about it, 2026
  3. Gmiterko, D., “Framing for actionable lint reports” at GitHub - Built for developers, March 2026 eslint-formatter-actionable ▷
  4. Gmiterko, D., “On ensurance of coding patterns” at GitHub - Where the world builds software, March 2026 eslint-plugin-defensive-coding ▷
  5. Gmiterko, D., “Dependency subgraphs and their application in image building” at GitHub - Build software better, together, Jun 2026 yarn-plugin-prod-lockfiles ▷
ToolingValidationESLint
Made by

DeeDominik Gmiterko

References

LinkedIn
@dee.gmiterko
Github
m.me/dee.gmiterko

Huh, and how is it going?
We will have to see…
But for now I could tell you a bit more about myself. For starters, I've always been an artist. One way or another.

Contact

You can start your very own conversation with me...
Feel free to reach out to me at your preferred platform 😊
Dominik Gmiterko
Email
me@deegmiterko.com
Messenger
m.me/dee.gmiterko
Instagram
@dee.gmiterko
LinkedIn
dominik-gmiterko
Discord
dee.gmiterko
© 2026 Dominik Gmiterko