What I'm building, learning, and focused on right now.
A snapshot of what I'm building, learning, and focused on right now. Less about who I am, more about what has my attention today.
Last updated July 2026
Building
- Full-stack and AI products
Shipping production features across e-commerce, Web3, and applied AI systems.
A drop-in chat widget that answers your visitors from your own pages, with citations, never made up. Point it at URLs, a sitemap, or PDFs; theme it to your brand; bring your own model (OpenAI, Anthropic, Gemini) or use mine. Next.js + Neon pgvector + Gemini, with GitHub/Google/Discord sign-in.
A full Medusa storefront where you shop by describing what you need instead of guessing keywords (pgvector and Gemini), with cart, Stripe checkout, customer accounts, and transactional email. Open source.
A free bulk image compressor that runs entirely in your browser, nothing is ever uploaded. WebP, JPEG, and PNG with lossy quantization, plus an Auto mode that picks the smallest format per image. Built on WebAssembly codecs in a web worker.
A free cross-platform desktop app to download YouTube video, audio, and playlists on your own machine, with subtitles, SponsorBlock, trimming, and auto-update. An Electron shell over yt-dlp and ffmpeg, for Windows, macOS, and Linux.
The Ask AI chat does things, not just answers: jump to a section, open a post, filter my projects by tech, draft a message or email to me, or switch the theme. It runs free on Gemini with a five-model fallback chain, answers detailed questions straight from my blog posts (RAG), and takes voice input and reads its answers back on mobile.
The blog search finds posts by meaning, not just keywords, reusing the same embeddings as the assistant.
- Live chat bridged to Telegram
Visitors reach me in real time; messages land in my Telegram and replies stream back to the widget.
Exploring
- Agentic coding workflows
Going deeper with Claude Code, Cursor, and Codex as part of the daily loop.
- Tool-calling LLMs in the browser
Client-side tools with the Vercel AI SDK, where the model decides and the browser acts.
Shipping next
- Answer-aware follow-ups
Suggesting the next question from the conversation itself, instead of a fixed list.
Packaging this site's assistant into an embeddable widget anyone can drop into their own site, docs, or store, grounded in their own content. Try the building blocks in the playground.
Writing
Notes on RAG, agents, commerce, and Web3 from real projects.