BLK Exchange
A financial literacy game dressed up as a stock market, where the 36 companies you trade are all from the Black economy.

01
The Problem
For every $100 in wealth held by White families, Black families hold $15. Black Americans answer 37 percent of financial literacy questions correctly against 55 percent for White Americans, and the gap survives controlling for education. Only 18 percent of Black households own stocks. The tools that exist to fix this trade Apple and Tesla on Bloomberg headlines, which assumes a familiarity with the market the intended audience disproportionately does not have. The information is getting through, because people watch Earn Your Leisure and follow the finance accounts. What is missing is anywhere to practice in a context that feels like their own.
02
The AI Design Decision
I used two AI models for two different jobs rather than one for both. A fast one sorts incoming news all day, because there's a lot of it and speed is the whole requirement. A slower, better one writes the market events players actually read, because that's the writing they'll judge the game on. What I chose not to build is a market that just goes up. That's what a generative model gives you by default, and it teaches the one lesson this audience can least afford.
03
The Architecture
- 01Watch 16 Black media publications for real news, all day
- 02sort each story by which company and which lesson it touches
- 03a queue decides what the market does next, so every player meets all 23 concepts
- 04the game writes the news event and moves the price
- 05a check on the last few events forces the mood back the other way if it has drifted too positive
04
The Responsible AI Constraint
Ask a language model to write market news and it's relentlessly cheerful. A simulator that only goes up teaches the one lesson this audience can least afford, that investing is a machine which always pays out. So the balance isn't a line in a prompt asking the model to be fair. The game counts what it actually produced. It looks back at the recent run of events, works out how positive they've been, and if the mood has drifted too far one way it forces the next one back the other. It waits until there are five events before it starts, because you can't measure a trend in two. And events that moved no prices don't count toward the tally, so the correction responds to what players actually felt.
05
The Result
Winner of the Design Award at Hackonomics 2026, and live at blkexchange.com. Built alone in one week with Claude Code by someone who isn't a software engineer: 36 companies across 12 industries, 18 linked tables of game data, a news reader watching 16 publications, and 23 money concepts in a section players unlock as they go.
06
If you remember one thing
Ask a model to write market news and it's relentlessly cheerful. A game about money that only goes up teaches the worst possible lesson, so the game counts its own mood and forces it back the other way.
Tech Stack
- Next.js 14 (App Router)
- React 18
- Convex
- Clerk
- Claude Sonnet 4.6
- Groq (llama-3.3-70b)
- ElevenLabs Flash v2.5
- Perplexity Search API
- Firecrawl
- Finnhub
Links