name 100 games

name 100 games, out loud, fastest time wins.

Comment

Inspiration

I wanted to create a fun challenge and to try and push the limits of voice detection. It started as a voice-controlled library import experiment and the naming part turned out to be the fun part.

What it does

You name 100 video games out loud, against the clock. The hard part is judging your answer instantly... It uses a combination of browser dictation and layers of caching, databases, API calls to try and validate the users answer as fast as possible while maintaining accuracy.

Ludwig enjoying an early iteration of the challenge

How I built it

Heavy use of codex 5.6 and a lot of iteration. Local-first judging so the browser answers before any network call, layered fallbacks (cache → DB → ChatGPT 5.6 → API), a ~163k-title judge catalog, and the self-learning loop where unresolved titles get picked up on first IGDB touch.

The ChatGPT call happens in async (to reduce latency of feedback), providing an overruling in the most difficult cases and teaches aliases.

Challenges I ran into

A big challenge continues to be accents, speech patterns, and different title variations. One big challenge was massive user spike events resulting in API limit bottlenecks and the Vercel/Neon usage limits hitting the cap of our plan (both fixed now) causing degradation of judging performance.

The alias learner has logged over 1,000 misheard phrases in two weeks: things like "Palaworld" for Palworld and "Bad company too" for Battlefield: Bad Company 2

Accomplishments that I'm proud of

Thousands of unique visitors this week with lots of invested players having lots of fun (and some very frustrated ones).

  • 3,000+ runs started in the last two weeks, and 443 people actually made it to 100 (it's harder than it sounds)
  • 176 players finished a run and immediately started another one
  • the judge catalog has grown to ~167k titles with aliases, and it keeps teaching itself new ones from what players say
  • 97.8% of answers get judged entirely in the browser — the 5,000 most-named titles ship as a 61 KiB pack, so most verdicts never touch a server

What I learned

Unbelievable amounts, probably the most valuable project I've ever done learning wise.

  • design philosophy (what makes a compelling user experience)
  • good UI design (work in progress)
  • SEO (AI SEO?)
  • 0 budget marketing -> driving users to the product
  • building FAST
  • strong understanding of agentic development
  • improved cost efficient decision making and planning in project

What's next for name 100 games

  • Improving speed, accuracy, and the self learning mechanism
  • Training a small, auditable reranker using string, phonetic, popularity, provenance, and ambiguity features
  • Bringing joy to more users

Built With

Share this project:

Updates