A mix of client work, contributions landed upstream, and side projects that outlived a weekend. Filter by technology if that's useful.
Auto-generated, type-safe Telegram Bot API client for Go. Tracks the upstream Bot API v10.0 spec — 176 methods, 301 types, 1428 generated tests — so the library never lags behind Telegram releases. Batteries-included: long-polling, retries, context cancellation, structured errors. MIT licensed, zero hand-written API surface, fully tested via codegen.
Patent-pending vector database that stores no vectors. Instead of persisting embeddings, GraphANN navigates a graph over compressed source text and recomputes embeddings on demand for the handful of nodes it actually visits — typically 200–500 per query. Result: ~97% less storage than a conventional vector DB, recall on par with a flat index, and hot-swappable embedding models with zero re-indexing. Ships as a single zero-dependency binary with multi-tenant isolation, zone-aware HA clustering, and pluggable inference backends (Ollama, ONNX, OpenAI, custom). A storage-efficient replacement for the usual vector database tax at production scale.
Developer productivity tool that transforms GitHub activity into a gamified experience. Analyses contributions and generates interactive dashboards with 115+ achievements across 26 categories. Features team analytics, GitHub Action integration, smart caching, and bot filtering.
Persistent memory system for Claude Code that enables long-term context retention across coding sessions. Enhances AI-assisted development by maintaining project-specific knowledge and conversation history.
kportal is a Kubernetes port forwarding manager that allows you to define all port forwarding rules for your development project, fire and forget. It keeps the connections alive with active monitoring, has relatively cute and simple terminal UI and does what it says on the tin. No, it's not a wrapper for the kubectl.
Production-ready Kubernetes controller for automatically mirroring resources across namespaces. Unlike alternatives, works with any resource type including Secrets, ConfigMaps, Ingresses, and custom resources. Features environment-specific transformations, intelligent change detection with SHA256 hashing, and 90% reduction in API server load through optimised filtering. Tested managing 1000+ mirrors across 200+ namespaces with minimal memory footprint.
Software delivery methodology focused on balanced, empowered, async teams (BEAT). A framework for organising development workflows that emphasises team autonomy, asynchronous communication, and sustainable delivery practices.
Plugin for the Traefik load balancer allowing for OIDC authentication of the specific routes. It's writen from scratch with minimal dependencies to allow for easier auditing.
GraphQL monitoring proxy for monitoring and logging GraphQL queries. Created as a form of protest against Hasura pricing model. Evolved into a full-fledged proxy with support for multiple backends, query validation, dynamic cache and more.
Terminal-based utility for managing /etc/hosts file with interactive TUI. Features group-based categorisation, automatic backups, secure daemon architecture using Unix sockets, and domain protection. Works without sudo for regular users.
Kubernetes operator for managing Kubernetes Jobs. It allows for creating ordered jobs in a declarative way for example for data processing and Machine Learning pipelines. It's written in Go as well.
Go library for running GraphQL queries with minimal overhead written in Go. The concept was simple - copy / paste query from GraphQL playground and run it in Go. It's used in multiple projects and is a part of the telegram-bot.app stack.
Semantic version generator which can be used as a Github Action. Figures out the semantic version calculation methods for every commit, tag or pull request.
24 Raspberry Pi HA Kubernetes cluster on K3S, running Coral TPU, Longhorn, Victoria Metrics stack and so on. Internet via WAN balancing and wireguard with three independent providers. Website you're currently viewing is hosted on it.
Fastest and smartest bot for Telegram groups management. Built in Go, using NATS for communication between microservices and GraphQL + PostgreSQL for database queries. Additional features include Python + tensorflow for image recognition ( trained and running in-house ), content classification and sentiment analysis. Frontend built with Quasar Framework and VueJS.