ainews247.org
AiNews247
Visit →A daily AI news surface. Pulls from Hacker News, Reddit, TechCrunch and the rest, then dedupes and ranks it all so the signal sits on one page instead of across ten tabs.
Product-minded builder. Customer-obsessed. Dangerously fast at shipping AI systems that drive revenue. 10+ years professional experience, coding since 5th grade.
Deep empathy for customer pain. Continuous discovery. Validate before build. Every line of code justified by real user needs.
Ship working solutions in days, not months. YAGNI practitioner. No over-engineering. Iterate towards product-market fit.
Agentic AI, RAG, multi-LLM routing at scale. Harden systems only when validated by customer pain, not vanity metrics.
$50M ARR impact through business-first engineering. Every technical decision tied to customer value and business outcomes.
I build hyperlocal grounding systems—RAG architectures that ground LLMs with location-specific business data at scale. My systems serve brands like AT&T, Walmart, Jersey Mike's, and Western Union across 1,000+ locations.
ainews247.org
A daily AI news surface. Pulls from Hacker News, Reddit, TechCrunch and the rest, then dedupes and ranks it all so the signal sits on one page instead of across ten tabs.
domsy.io
A prototyping companion that turns a rough idea into a working interface. Conversation on one side, live app on the other — built for the ten minutes before you commit to anything.
orbiter-mods.com
Mod hosting and the Orb installer for Orbiter Space Flight Simulator. Browse add-ons, download them, and get them into the game without the usual zip-file archaeology.
riter.cc
An AI writing workforce — specialised agents that research, draft and revise in parallel. Writers direct the work instead of babysitting a single chat window.
z — terminal harness
An agentic coding harness for the terminal. Bring your own keys, hot-swap between 26 providers mid-session with /providers use, keep context alive across compaction, and drive the whole thing from your phone over Telegram.
dlgo — local inference
LLM inference in pure Go. Runs GGUF models on CPU or Vulkan with no Python, no CUDA runtime and no external inference server — one binary that streams tokens and reports honest tok/s and TTFT. That is the real UI above, 10.7 tok/s on a laptop CPU.
framora — desktop editor
A hybrid live-preview Markdown editor for Windows, macOS and Linux. Ghost-tag editing renders syntax in place as you type, 1,000 templates ship in the box, and the AI assistant runs entirely on dlgo — no cloud, no API key.
Cut through GenAI hype. Ship AI products that customers actually want. Business-first engineering that drives revenue.
Book a CallBuild working prototypes in days, not months. Continuous discovery. Pragmatic architecture. Only build what customers need.
Schedule Time