Active

On Dima's Radar

Atlanta events ranked against my actual taste, then turned into easy invitations that help me do more of what brings me joy and stay closer to people I care about.

Used to send real invitations and turn reconnections into specific plansOne invite can combine events, open dates, and evergreen fallbacksNo formal response-rate or event-coverage measurement yet

The Problem

I kept finding out about great Atlanta stuff after it happened. Eventbrite could give me 500 events, but it couldn’t answer the question I actually had: I want to see Dave — what are six things we’d both like, and which three nights am I free?

I wanted more music, dancing, outdoor adventures, and family activities. With two small kids, calendars did most of the deciding. The hard part was turning “we should hang out” into a plan someone could say yes to.

What It Is

Radar scrapes the city wide, then ranks everything against a taste profile built from years of my concert history and Spotify data — each event scored 0–100 by an LLM across five axes: genre match, venue quality, format rarity, lineup strength, value for money. Radar charts show why each event received its score. A wizard walks through a few preference questions and re-ranks everything live.

The feed updates continuously; alongside dated events, a deep bench of evergreen activities (free hikes, pools, family spots) covers the “what do we do today” case.

On Dima's Radar showing recommended Atlanta events
The feed is deliberately broad. The useful output is deliberately short.

The Part That Matters: The Invite

Finding the right event is one half of the job. The other is turning it into a plan: pick a handful of events plus a few free evenings, and send one link or a plain-text invite a friend can answer with a yes or a maybe in a single text. Proposed times and evergreen fallbacks reduce the back-and-forth that usually kills a plan.

When It Became Real

By July, I was no longer only demoing Radar. I was using it in the part of life it was built for: reconnecting with people.

In one conversation, a friend and I moved from talking generally about getting together to a short invitation with a Shakespeare performance, a storytelling night, live music, and a specific afternoon for coffee. In coaching, I caught myself describing the site as something I had been using to make plans and connections. That mattered more than merely having built it.

The outcome I care about is a plan that leaves the group chat.

How I Use It

  • Weekend scan — Friday morning, see what’s good in the next 2–3 days
  • Spontaneous Sunday — browse evergreen activities for impromptu outings
  • Date night — filter for “Dima + Jeannie, no kids”
  • Family outing — age-gated activities filtered for a 4-year-old
  • Reconnection — a weekly script pairs overdue friends with upcoming events as a concrete excuse to reach out

What Is Still Uncertain

  • Whether people besides me find the ranking meaningfully better than a normal calendar
  • Which sources and neighborhoods the scraper still misses
  • Whether an invitation link gets a faster answer than a thoughtful text
  • How much calendar context can be useful without making the tool invasive

Those are usage questions, not feature requests. The next version should come from watching people plan.

Models, agents, tools, and infrastructure

Claude + taste profiles

Scores a wide event feed against my actual music, venue, format, and value preferences.

Scraping + public event feeds

Pulls from many sources because no single Atlanta calendar covers the city well.

Vanilla HTML, CSS, and JavaScript

Keeps the event browser, filters, maps, and invitation flow fast and portable.

Leaflet

Maps events and evergreen activities with clustered, mobile-friendly markers.

Calendar context

Turns event discovery into a practical choice among nights that could actually work.

Vercel analytics + PostHog

Tracks how the live site gets used without pretending usage already proves impact.

Up next

All Roads to Kill Bill
All Roads to Kill Bill
An interactive 3D influence network mapping every film that fed into Kill Bill — and every film it spawned. 200+ nodes, live force simulation. Built with React, Three.js, and TMDB.