lab463

About — English

lab463 is the personal lab and portfolio of Martin Horný, a .NET developer based in Košice, Slovakia. He builds tools for problems he actually has, then ships and uses them. Ten projects are currently running and in active use.

Projects (9 of 9 shipped & running)

  1. SignalFinder — Trend intelligence prototype. URL to clean Markdown plus external reference graph; Gemini extracts topics and claims; Serper validates trends across five regions (US, DE, SK, UK, FR). Built more as a services-integration experiment than a finished tool. Stack: .NET 8, React, Vite, PostgreSQL, Gemini, Serper.
  2. VinneGPT — RAG-based AI assistant about the village of Vinné, Slovak NLP, source-cited answers, Gemini embeddings + Pinecone vector search. Stack: .NET 10, React.
  3. InvoiceParser — Email-in / XML-out invoice parser using Claude Haiku as an end-to-end vision pipeline (no separate OCR). SK/CZ accounting compatible.
  4. LinkVault — Bookmark manager. PWA + Chrome extension (popup + sidepanel). Data stored in the user's own Google Drive. Real-time sync via SignalR. Offline-first via Dexie. Reader mode via Mozilla Readability.
  5. BillSharing — Split a bill from a receipt photo. Claude vision reads items, SignalR syncs selections across devices, Pay-by-Square QR per person. No install, no registration.
  6. SpecTacular — SPEC-driven development workflow with a Visual Studio plugin for in-IDE markdown spec editing and step-by-step execution state.
  7. claude-ideas — Open-source slash-command toolkit for Claude Code; project-aware variants for .NET and React; agent caching; reflective skill improvement. Version 4.9.
  8. SchoolAI — Public demo dashboard for an AI manager role in a school: accounts, access, activity, costs.
  9. HomePanel — Physical smart-home control: 10 panels, 4 buttons + 4 LEDs each, ESP32 + Tasmota32, RS485 Modbus, MQTT to Home Assistant, bidirectional LED feedback. Hardware schematic co-designed with Claude.

Daily-driver stack

Backend: .NET 10, C#, SQL Server, Entity Framework. Frontend: React, TypeScript, Vite, Tailwind, Zustand, Dexie. Real-time: SignalR. AI: Claude (Haiku, Sonnet), Gemini. Vector: Pinecone, Gemini multilingual embeddings. Browser: Chrome Extensions API, CRXJS, Service Workers. Infra: Docker, Nginx, Cloudflare Tunnel, self-hosted. Embedded: ESP32, Tasmota32, Berry, RS485/Modbus, MQTT, Home Assistant. Tooling: Git, Azure DevOps, Claude Code.

Writing

Contact

GitHub: github.com/Tadzesi. LinkedIn: linkedin.com/in/martin-horny-profile. Email: martin (at) lab463.com.

O projekte — slovenčina

lab463 je osobné laboratórium a portfólio Martina Horného, .NET developera z Košíc. Stavia nástroje pre problémy ktoré reálne má, dokončí ich a používa. Aktuálne beží desať projektov.

Projekty (9 z 9 nasadených a bežiacich)

  1. SignalFinder — Prototyp pre trend intelligence. URL prevedie na čistý Markdown a graf externých odkazov; Gemini identifikuje témy a tvrdenia; Serper overuje trendy v piatich regiónoch (US, DE, SK, UK, FR). Skôr experiment so spojením služieb než hotový nástroj. Stack: .NET 8, React, Vite, PostgreSQL, Gemini, Serper.
  2. VinneGPT — AI asistent o obci Vinné, slovenské NLP, odpovede s citáciou zdroja, Gemini embeddings + Pinecone. Stack: .NET 10, React.
  3. InvoiceParser — Faktúra emailom, XML späť. Claude Haiku ako end-to-end vision pipeline (žiadny samostatný OCR krok). Kompatibilné so SK/CZ účtovníctvom.
  4. LinkVault — Bookmark manager. PWA + Chrome rozšírenie. Dáta v používateľovom Google Drive. Real-time sync cez SignalR. Offline-first (Dexie). Reader mode (Mozilla Readability).
  5. BillSharing — Rozdelenie účtu z fotky bloček. Claude vision rozozná položky, SignalR synchronizuje výber, Pay-by-Square QR pre každého. Bez inštalácie, bez registrácie.
  6. SpecTacular — SPEC-driven development. Visual Studio plugin pre editovanie markdown špecifikácie priamo v IDE.
  7. claude-ideas — Open-source toolkit slash commandov pre Claude Code; varianty pre .NET a React; agent caching. Verzia 4.9.
  8. SchoolAI — Verejné demo dashboardu pre AI manažéra v škole: účty, prístupy, aktivita, náklady.
  9. HomePanel — Fyzické ovládanie smart home. 10 panelov, 4 tlačidlá + 4 LED, ESP32 + Tasmota32, RS485 Modbus, MQTT do Home Assistant. HW schému navrhol Claude.

Denný stack

Backend: .NET 10, C#, SQL Server, Entity Framework. Frontend: React, TypeScript, Vite, Tailwind, Zustand, Dexie. Real-time: SignalR. AI: Claude (Haiku, Sonnet), Gemini. Vector: Pinecone. Browser: Chrome Extensions API, CRXJS. Infra: Docker, Nginx, Cloudflare Tunnel, self-hosted. Embedded: ESP32, Tasmota32, RS485/Modbus, MQTT, Home Assistant.

Kontakt

GitHub: github.com/Tadzesi. LinkedIn: linkedin.com/in/martin-horny-profile. Email: martin (at) lab463.com.

Martin Horný

.NET developer · Košice, SlovakiaKošice, Slovensko

I build tools for problems I actually have. Then I ship them. Then I use them.

10 projects. All running. All used. From RS485 wiring to React PWA to Chrome extension — whatever the problem needs.

Staviam nástroje pre problémy ktoré reálne mám. Potom ich dokončím. Potom ich používam.

10 projektov. Všetky bežia. Všetky používam. Od RS485 zapojenia po React PWA po Chrome rozšírenie — podľa toho čo problém vyžaduje.

§ 01–08 Projects · shipped & running Projekty · nasadené a bežiace 9 / 9
01signalFinder
AI · trend intelligence · prototype AI · trend intelligence · prototyp

SignalFinder

Trend intelligence prototype — services stitched together.

Prototyp pre trend intelligence — služby pospájané dokopy.

Paste any article URL, get clean Markdown plus a graph of external references. Gemini extracts topics and claims; Serper validates trends across five regions (US, DE, SK, UK, FR). Phase 1 — extraction and reference graph — ships and runs end-to-end; Phase 2 (cross-region validation) is in progress.

Cez URL extrahuje článok, prevedie ho na čistý Markdown a vytiahne graf externých odkazov. Gemini identifikuje témy a tvrdenia, Serper overuje trendy v piatich regiónoch (US, DE, SK, UK, FR). Phase 1 — extrakcia a graf referencií — beží end-to-end; Phase 2 (cross-region validácia) je rozrobená.

Built more as a test of how cleanly Gemini, Serper and a custom .NET extraction pipeline compose than as a real product. The output works, but the value is the integration lesson — not yet a tool.

Postavené skôr ako test toho, ako čisto sa Gemini, Serper a vlastná .NET extrakcia spoja dokopy, než ako reálny nástroj. Výstup funguje, ale skutočná hodnota je v tej integrácii — zatiaľ to nie je produkt.

// technical detail
extraction
.NET 8 pipeline → HTML to Markdown, language detection, external link graph
.NET 8 pipeline → HTML do Markdownu, detekcia jazyka, graf externých odkazov
analysis
Gemini for topics, claims, relevance scoring
validation
Serper search API → 5 regions (US · DE · SK · UK · FR)
storage
PostgreSQL · EF Core migrations
stack
.NET 8 · React · Vite · PostgreSQL · Gemini · Serper
infra
Docker · self-hosted · Cloudflare Tunnel
Open SignalFinder↗ Otvoriť SignalFinder↗
02vinneGPT
AI · RAG · Slovak NLP AI · RAG · slovenské NLP

VinneGPT

AI assistant for the village of Vinné.

AI asistent o obci Vinné.

The village of Vinné has public information scattered across its website. A custom scraper pulls and indexes it weekly into a vector database. VinneGPT answers questions in Slovak with source citations — no hallucinations, only what's in the sources.

Obec Vinné má verejne dostupné informácie rozsypané po webe. Vlastný scraper ich každý týždeň stiahne a indexuje do vektorovej databázy. VinneGPT odpovedá na otázky v slovenčine s citáciou zdroja — žiadne halucinácie, iba to čo je v zdrojoch.

Built as a hands-on experiment with RAG and vector search. I now apply the same patterns at work for auto-generating code documentation on merge requests.

Vzniklo ako experiment s RAG a vector searchom. Tieto skúsenosti teraz používam v práci pri automatickom generovaní dokumentácie z kódu pri merge requestoch.

// technical detail
scraper
custom .NET background job, weekly cycle, recursive crawl
vlastný .NET background job, týždenný cyklus, rekurzívny crawl
chunking
optimized for Slovak text
optimalizovaný pre slovenský text
embeddings
Gemini multilingual → Pinecone vector search
stack
.NET 10 · React · Gemini API · Pinecone
infra
Docker · self-hosted · Cloudflare Tunnel
Open VinneGPT ↗ Otvoriť VinneGPT ↗
03invoiceParser
Vision LLM · automation Vision LLM · automatizácia

InvoiceParser

Invoice by email → structured XML back.

Faktúra emailom → XML späť.

Forward an invoice by email. The backend catches it and sends the attachment (PDF, scan, photo) directly into Claude Haiku — no separate OCR step, the model handles images and text natively. The result is structured XML with extracted fields including company IDs, line items, VAT, due dates — compatible with SK/CZ accounting systems. XML comes back by email.

Prepošlite faktúru emailom. Backend ju zachytí, pošle prílohu (PDF, scan, fotka) priamo do Claude Haiku — žiadny samostatný OCR krok, model spracuje obrázok aj text natívne. Výsledok je štruktúrované XML s extrahovanými poliami vrátane IČO, DIČ, položiek, DPH, dátumov splatnosti — kompatibilné so SK/CZ účtovnými systémami. XML príde späť emailom.

The interesting part: Claude Haiku as an end-to-end vision pipeline replaces the entire OCR layer. Works on scans, photos, and digital PDFs without code changes.

Zaujímavá časť: Claude Haiku ako end-to-end vision pipeline nahrádza celú OCR vrstvu. Funguje na scany, fotky aj digitálne PDF bez zmeny kódu.

// technical detail
stack
.NET 10 · Claude Haiku (vision) · email backend
input
PDF · scanned documents · invoice photos
PDF · skenované dokumenty · fotky faktúr
output
structured XML · SK/CZ accounting compatible
štruktúrované XML · kompatibilné so SK/CZ účtovníctvom
infra
Docker · self-hosted
InvoiceParser↗ How it works→ Ako to funguje→
04linkVault
PWA · Chrome extension · real-time sync PWA · Chrome rozšírenie · real-time sync

LinkVault

Bookmark manager. Your data, your Google Drive.

Bookmark manager. Vaše dáta, váš Google Drive.

Raindrop and Pocket lock your data in their cloud. LinkVault stores everything in your Google Drive — exportable JSON, full ownership. PWA, Chrome extension popup and sidepanel stay in sync in real time via SignalR: save a bookmark in the extension and the PWA sees it instantly.

Raindrop a Pocket uzamknú vaše dáta v ich cloude. LinkVault ukladá všetko do vášho Google Drive — exportovateľné JSON, plná kontrola. PWA, Chrome extension popup aj sidepanel sú synchronizované v reálnom čase cez SignalR: uložíte bookmark v extension a PWA ho vidí okamžite.

Reader mode (Mozilla Readability) fetches article content on save. Works offline — articles readable without internet. Sync conflicts use merge logic, not last-write-wins.

Reader mode (Mozilla Readability) stiahne obsah článku pri uložení. Funguje offline — články čitateľné aj bez internetu. Pri konfliktoch medzi klientmi sa nerobí jednoduché last-write-wins: merge logika zachová zmeny z oboch strán.

// technical detail
frontend
React · TypeScript · Vite · Tailwind · Zustand · Dexie (offline-first)
extension
CRXJS · Chrome APIs (popup + sidepanel + background worker)
backend
.NET 10 · SignalR · Google OAuth · Google Drive API
infra
Docker · self-hosted
Open PWA↗ Otvoriť PWA↗ Chrome Web Store↗ About→ O projekte→
05billSharing
vision OCR · multi-device sync · zero install vision OCR · multi-device sync · bez inštalácie

BillSharing

Split a bill from a photo. QR code. Done.

Rozdelenie účtu z fotky bloček. QR kód. Hotovo.

Started as a quick Claude HTML page at the pub: photo of the bill, add colleagues, everyone taps what they had on my phone, everyone pays manually. It worked, but didn't scale.

Vzniklo pri pive ako jednoduchá Claude HTML stránka: odfotiť blok, pridať kolegov, každý si v mojom telefóne vykliká čo jedol, každý zaplatí manuálne. Fungovalo, ale neškálovalo.

Now: photo of the receipt → Claude reads the items → add your bank account → share a QR with colleagues. Each person opens it on their own phone, taps their items (SignalR syncs selections in real time across all devices) → gets a payment QR directly in their banking app with the exact amount.

Teraz: odfotíte účtenku → Claude rozozná položky → pridáte číslo účtu → zdieľate QR kód kolegom. Každý otvorí na svojom telefóne, vykliká položky (SignalR synchronizuje výber v reálnom čase medzi všetkými zariadeniami) → dostane QR kód priamo do svojej bankovej aplikácie s presnou sumou.

No registration. No app to install. Just a QR.

Žiadna registrácia. Žiadna aplikácia na inštaláciu. Iba QR.

// technical detail
stack
.NET 10 · SignalR · Claude vision (receipt OCR)
payment QR
Pay by Square (SK standard)
Pay by Square (SK štandard)
infra
Docker · self-hosted
Try BillSharing↗ Vyskúšať BillSharing↗
06specTacular
developer tooling · VS plugin · workflow vývojárske nástroje · VS plugin · workflow

SpecTacular

SPEC-driven development. Understand first, code second.

SPEC-driven development. Najprv porozumenie, potom kód.

The common problem with AI and code: you start generating before both you and the AI actually understand what's being built. The result is hallucinations, rewrites, lost context.

Bežný problém s AI a kódom: začnete generovať skôr než obaja (vy aj AI) skutočne rozumiete čo sa stavia. Výsledkom sú halucinácie, prepísané časti, stratený kontext.

SpecTacular reverses this: a structured conversation with AI about the project — from basic functionality to tech stack — until a complete specification exists. A VS plugin enables in-IDE editing of the MD spec and shows execution state step by step. Input is a project description. Output is a working application or service without hallucination.

SpecTacular to obracia: postupný rozhovor s AI o projekte — od základnej funkcionality po tech stack — kým nevznikne kompletná špecifikácia. VS plugin umožňuje editovanie MD špecifikácie priamo v IDE a pri exekúcii zobrazuje stav jednotlivých krokov. Vstup je popis projektu. Výstup je funkčná aplikácia alebo služba bez halucinovania.

I use it on my own projects. Colleagues and friends use it too.

Používam na vlastných projektoch, používajú kolegovia a kamaráti.

GitHub → SpecTacular↗ Documentation↗
07claude-ideas
prompt engineering · open source · v4.9 prompt engineering · open source · v4.9

claude-ideas

Slash commands for Claude Code. Prompt engineering toolkit.

Slash commandy pre Claude Code. Prompt engineering toolkit.

A collection of Claude Code slash commands for prompt engineering, technical analysis, content generation, and deployment workflows. Project-aware variants for .NET and React automatically scan project structure before responding. Agent caching speeds up repeated prompts 10–20×. The /reflect command analyzes sessions and proposes skill improvements from feedback.

Kolekcia Claude Code slash commandov pre prompt engineering, technickú analýzu, generovanie obsahu a deployment workflow. Project-aware varianty pre .NET a React automaticky skenujú projektovú štruktúru pred odpoveďou. Agent caching zrýchľuje opakované prompty 10–20×. Systém /reflect analyzuje session a navrhuje vylepšenia na základe spätnej väzby.

Version 4.9 (April 2026). 11 commands. PowerShell installer. VitePress documentation.

Verzia 4.9 (apríl 2026). 11 commandov. PowerShell installer. VitePress dokumentácia.

GitHub → claude-ideas↗ Documentation↗ Dokumentácia↗
08schoolAI
demo · dashboard · education demo · dashboard · školstvo

SchoolAI

Dashboard demo for an AI manager in a school setting.

Demo dashboardu pre AI manažéra v prostredí školy.

A live demo: one place where a school’s AI lead can see how AI is being used across the building — accounts, access, activity, costs. Built to make the role of “AI manager” concrete, not abstract: what does a person responsible for AI in a school actually need to look at every week?

Bežiace demo: jedno miesto, kde AI manažér školy vidí, ako sa AI využíva naprieč školou — účty, prístupy, aktivita, náklady. Postavené aby zhmotnilo rolu „AI manažéra“: na čo sa človek zodpovedný za AI v škole reálne potrebuje pozerať každý týždeň?

Demo only — fictional data, real UI. Open the link, click around.

Iba demo — fiktívne dáta, reálne UI. Otvorte odkaz, vyskúšajte si to.

// technical detail
stack
React · TypeScript · .NET 10
audience
cieľová skupina
school principals · IT coordinators · AI leads
riaditeľov škôl · IT koordinátori · AI vedenia
status
public demo · not in production
verejné demo · nie v produkcii
infra
Docker · self-hosted
Open SchoolAI demo↗ Otvoriť SchoolAI demo↗
09homePanel
ESP32 · Tasmota · Home Assistant

HomePanel

Physical smart home panel. Claude designed the circuit. I built it.

Fyzický smart home panel. Claude navrhol schému. Ja som to postavil.

10 physical button panels throughout the house, each with 4 buttons and 4 LED indicators. ESP32 running Tasmota32 reads inputs via RS485 Modbus, sends events to Home Assistant via MQTT, and controls a 16-channel relay board for lights. LED feedback is bidirectional — when someone turns on a light by voice or app, every physical panel LED updates instantly.

10 fyzických panelov po celom dome, každý so 4 tlačidlami a 4 LED. ESP32 s Tasmota32 číta vstupy cez RS485 Modbus, posiela udalosti do Home Assistant cez MQTT, riadi 16-kanálové relé. LED spätná väzba je obojsmerná — keď niekto zapne svetlo hlasom alebo appkou, každá fyzická LED sa okamžite aktualizuje.

HW schematic designed with Claude. Built, flashed, and running.

HW schému navrhol Claude. Postavené, oflashované, beží.

// technical detail
controller
ESP32 DevKit · Tasmota32 · Berry scripting
io
23IOE48_NPN · 40× input · 40× LED output · RS485 Modbus RTU
relay
Waveshare POE Relay 16CH · Modbus TCP
motion
4× PIR · Home Assistant binary_sensor
click
SINGLE / DOUBLE / HOLD per button
10kitchenPanel
.NET 10 · React · Raspberry Pi 4 · Home Assistant .NET 10 · React · Raspberry Pi 4 · Home Assistant

Smart Kitchen Panel

An ordinary TV in the kitchen that knows what is happening at home.

Obyčajný televízor v kuchyni, ktorý vie, čo sa v dome deje.

A portrait kitchen display driven by a TV remote: clock, weather, home status from Home Assistant, kitchen timers, cameras, a photo frame, and online radio and Spotify on a WiiM streamer. A motion sensor turns the TV off when nobody is in the kitchen and wakes it when someone walks in. Nothing was bought for it — the TV was a dumb one, the Raspberry Pi 4 used to run Home Assistant, and the panel itself runs on the home lab server.

Zvislý kuchynský displej ovládaný diaľkovým ovládačom od TV: hodiny, počasie, stav domu z Home Assistanta, kuchynské časovače, kamery, fotorámik a internetové rádio so Spotify cez WiiM. Pohybový senzor vypne televízor, keď v kuchyni nikto nie je, a zobudí ho, keď niekto vojde. Nič sa preň nekupovalo — televízor bol obyčajný, Raspberry Pi 4 kedysi bežal Home Assistant a samotný panel beží na domácom serveri.

The public page runs the real panel as a demo with fictional data. Anything that is not live is labelled as such.

Verejná stránka spúšťa skutočný panel ako demo s vymyslenými dátami. Všetko, čo nie je živé, je tak označené.

// technical detail
stack
.NET 10 · SignalR · React 19 · TypeScript · Vite · PostgreSQL 17
architecture
architektúra
Clean Architecture (Domain · Application · Infrastructure · Api · Web). One ASP.NET host serves the REST API, the SignalR hub and the built React app, so the whole panel is a single container.
Clean Architecture (Domain · Application · Infrastructure · Api · Web). Jeden ASP.NET host servíruje REST API, SignalR hub aj zbuildované React UI, takže celý panel je jeden kontajner.
real-time
real-time
SignalR pushes Home Assistant state changes to every screen the moment they happen. Presence arrives over a Server-Sent Events stream instead of the old 15 s polling.
SignalR posiela zmeny stavu z Home Assistanta na každú obrazovku hneď, ako nastanú. Prítomnosť prichádza cez Server-Sent Events namiesto pôvodného dopytovania každých 15 s.
presence
prítomnosť
Zigbee PIR sensor in Home Assistant. After 10 minutes of absence a watchdog on the Pi switches the HDMI output off (wlr-randr) and puts the TV to standby over HDMI-CEC. When someone returns it wakes the TV and calls POST /api/display/woke; the panel shows a wake flash, then Domov by day or the clock at night.
Zigbee PIR senzor v Home Assistantovi. Po 10 minútach bez pohybu watchdog na Pi vypne HDMI výstup (wlr-randr) a uspí TV cez HDMI-CEC. Keď sa niekto vráti, TV sa zobudí, zavolá sa POST /api/display/woke a panel ukáže záblesk prebudenia, potom Domov cez deň alebo hodiny v noci.
kiosk
kiosk
Raspberry Pi 4 · labwc (Wayland) · Chromium in kiosk mode · screen rotated in software (labwc output transform) because the TV hangs in portrait. A watchdog restarts Chromium after a crash and the SignalR client reconnects on its own.
Raspberry Pi 4 · labwc (Wayland) · Chromium v kiosk režime · obraz otočený softvérovo (transform výstupu v labwc), lebo TV visí na výšku. Watchdog reštartuje Chromium po páde a SignalR klient sa pripojí späť sám.
music
hudba
WiiM streamer over its HTTP API. Spotify cover art comes from getMetaInfo, radio track titles are read from the ICY metadata of the stream. A phone PWA is the same remote with a QR code on the panel. Fun facts about the music that is playing are generated only while someone is in the kitchen.
WiiM streamer cez jeho HTTP API. Obal albumu zo Spotify prichádza cez getMetaInfo, názvy skladieb z rádia sa čítajú z ICY metadát streamu. Mobilná PWA je ten istý ovládač s QR kódom na paneli. Zaujímavosti o hudbe, ktorá hrá, sa generujú len vtedy, keď je v kuchyni niekto prítomný.
photos
fotky
Google Photos Picker API: the app can only see the items the person picks on their phone, never the whole library. The QR code is drawn in the browser and never sent to a third party.
Google Photos Picker API: aplikácia vidí len položky, ktoré človek vyberie v telefóne, nikdy celú knižnicu. QR kód sa kreslí v prehliadači a nikdy sa neposiela tretej strane.
network
sieť
Local network only, TLS with a mkcert certificate, not routed through the Cloudflare Tunnel. Runs as a Docker stack (Node 22 + .NET 10 multi-stage build) on the home lab server.
Len lokálna sieť, TLS s certifikátom z mkcert, bez smerovania cez Cloudflare Tunnel. Beží ako Docker stack (multi-stage build Node 22 + .NET 10) na domácom serveri.
public demo
verejné demo
The same React app built with vite --mode demo: a virtual clock rebased to the visitor's wall clock, a fetch shim that answers /api from authored fixtures, and a fake SignalR hub. No request leaves the page for the home network.
Tá istá React aplikácia zbuildovaná cez vite --mode demo: virtuálne hodiny prispôsobené času návštevníka, fetch shim, ktorý odpovedá na /api z vlastných fixtures, a falošný SignalR hub. Žiadna požiadavka z tejto stránky nejde do domácej siete.
status
running at home · public demo
beží doma · verejné demo
Open the live demo↗ Otvoriť živé demo↗
+ more projects — internal tools, work experiments, things not ready for daylight yet. + ďalšie projekty — interné nástroje, firemné experimenty, veci ktoré ešte nie sú pripravené na svetlo sveta.
§ 09 Stack // daily-driver // denný driver
backend
.NET 10 · C# · SQL Server · Entity Framework
frontend
React · TypeScript · Vite · Tailwind · Zustand · Dexie
real-time
SignalR
ai / llm
Claude API (Haiku, Sonnet) · Gemini API
vector
Pinecone · Gemini multilingual embeddings
browser
Chrome Extensions API · CRXJS · Service Workers
infra
Docker · nginx · Cloudflare Tunnel · self-hosted
embedded
ESP32 · Tasmota32 · Berry · RS485/Modbus · MQTT · Home Assistant
tooling
Git · Azure DevOps · Claude Code
§ 10 Writing Texty 3 / 3

I write about things I build.

Píšem o veciach čo staviam.

§ 11 Contact Kontakt EOF

Let's build something.

Postavme niečo.

Martin Horný · .NET developer · Košice

Martin Horný · .NET developer · Košice

github
github.com/Tadzesi ↗
linkedin
/in/martin-horny-profile ↗
email
click to revealkliknite pre zobrazenie