I build tools that fix annoying or difficult problems: developer tooling, browser extensions, automation systems, security-focused software, and AI-assisted infrastructure. Most of what I make starts as a problem I want solved properly, then grows into something reusable.
-
JevSentinel (private during development; planned open-source release) — A self-hosted, context-aware moderation decision API being designed around TypeSafe Jev for games, forums, chat systems, bots, and online communities. Instead of acting as a simple profanity filter, JevSentinel is designed to evaluate harmful language together with intent, targeting, conversational context, roleplay, friendly banter, quoted/reported speech, severity, and optional moderation history. It then combines those probabilistic judgments with deterministic, operator-controlled policy and returns structured actions that the integrating application can enforce however it chooses.
The project is contract-first: its architecture, OpenAPI draft, configuration schema, policy/action model, privacy model, threat model, evaluation strategy, and AI-agent development rules are being defined before the provider implementation is finalized. It is also designed for custom actions, stateless deployments, configurable retention, human-review fallbacks, and conservative guardrails around high-impact automated moderation.
Planned stack: TypeScript, Node.js, Fastify, TypeSafe Jev, OpenAPI 3.1, JSON Schema, YAML
-
Paracci Secure Messaging A 100% offline, AI-resistant secure messaging application built around
.paraccienvelope files, hardware-calibrated Argon2id time-locks, and cryptographic sovereignty. Features a native desktop UI via Flask + pywebview, a self-inspecting integrity sentinel that triggers an automatic shutdown on tampering, native OS-level anti-screenshot protection, single-use burn tracking, and RAM-safe decrypted message handling. Source-available for independent security auditing. Stack: Python, Flask, pywebview, Cryptography (Ed25519, XChaCha20-Poly1305, Argon2id), Windows/Linux/macOS Native APIs -
YouTube Shorts Channel Blocker & Ultimate Downloader A full-featured quality-of-life extension I use daily. One-click channel blocking, automatic ad muting and skipping, Shorts shelf removal, and a dual-mode video downloader with a native yt-dlp companion app for 4K/1080p and MP3 downloads. Fully wired settings panel, real-time toggle integration, a native host setup page, and full 11-language UI support, all styled to match YouTube's exact design language. Stack: JavaScript, CSS, Chrome Extension V3 (MutationObserver, IntersectionObserver, Native Messaging)
-
X (Twitter) Auto Ad Blocker A lightweight Chrome extension I built for my daily browsing. Unlike traditional blockers that just hide elements with CSS, this extension mimics user interaction to permanently block the advertiser's account in the background. Because the account itself is blocked, your feed stays ad-free across all your devices. Also includes a media downloader for videos, GIFs, and images, and full 11-language UI support. Stack: JavaScript, CSS, Chrome Extension V3 (MutationObserver)
-
AITree - Project File Map Generator Scans any directory (or GitHub repo — no cloning required) and prints a clean, annotated file tree you can paste straight into any AI chat. Comes with a web UI, a live-watch mode, and a full MCP server so AI assistants like Claude can call it as a tool and read your files directly — without leaving the conversation. Stack: Python, MCP (Model Context Protocol), watchdog, tiktoken, gitpython
-
Facepunch Wiki Turkish Translation Project A large automated translation system that keeps Facepunch (Garry's Mod, Rust, Steamworks) wikis up to date in Turkish. Uses a triple-layer hybrid AI model (Google Gemini, DeepL, and local nllb-200) to process 8,000+ pages while protecting technical elements like code blocks, function names, and HTML structures during translation. Stack: Python, Node.js, Vanilla JS, Custom AI Validation
- Deep-Angler-Fish V2 (private) — A security research platform for internet asset discovery and external attack-surface observation. V2 is rebuilt around a modular control-plane foundation with passive certificate-transparency discovery, DNS/TCP/TLS/HTTP observation, tenant-scoped search and history, change detection and alerts, passive vulnerability-feed matching, regional launch controls, and governed network-location intelligence. Its contract-first, fail-closed architecture uses PostgreSQL as the authoritative source of truth, RabbitMQ for transient delivery, OpenSearch as a disposable projection, and Redis as a distributed rate authority, with tenant isolation, RBAC/RLS, audit context, idempotency, bounded retries, and cancellation built into the baseline. Stack: Python 3.12, FastAPI, PostgreSQL, RabbitMQ, OpenSearch, Redis, SQLAlchemy, Alembic, Docker
- Languages: JavaScript, TypeScript, Python, Go, HTML/CSS
- Backend & Tools: Node.js, Fastify, Express, FastAPI, Docker, OpenAPI
- Focus Areas: AI Decision Systems, Moderation Infrastructure, Developer Tools, Browser Extensions, Web Security Research, Open Standards