How a homeless 22-year-old built two AI apps to $1.5M revenue

Executive overview

Kelechi Onyeama moved from Nigeria to the US with $100, slept on a college floor, and was eventually made homeless by family before building Social Wizard — an AI dating-reply app — into nearly $1M in revenue within a year.

The key unlock was timing: he built alongside GPT-4 Vision's release, becoming first-to-market on a genuinely novel feature. A second app, CleanEats, followed and was acquired within months.

The fastest path to consumer app revenue is finding a feature that is inherently viral — something nobody has seen before — then distributing it through micro-creators before anyone else can copy it.

The three core human needs every successful app solves

1. Help people make more money
2. Help people find love
3. Help people have fun

(Health sits adjacent — it enables all three.)

Social Wizard: what made it work

• Built on GPT-4 Vision before the API was even out of rate-limited beta; launched with 10 API keys as a workaround
• First app to let users take a screenshot of an Instagram story and get contextualised reply suggestions
• Users generate ~10 hints per session but rarely copy one verbatim — the real product is a brainstorming partner, not a script
• Most common use case: asking a girl out; second most common: double-text anxiety
• A meaningful share of users are men already in relationships — unexpected finding from data

Onboarding: session one is everything

• Opens with a rigged test: the user tries to reply to a girl's message and is rated 1–5 by an LLM (presented as 1–10), guaranteeing a low score
• Seeing a "bad" score immediately shows the app's value without any explanation
• Follow-up questions prime for the paywall: tone preferences, communication goals
• Key metric tracked: time from install to first hint generated; volume of hints generated correlates with retention

Pricing

• Launched at $6.99/week; raised to $9.99–$10/week — LTV and revenue increased
• Weekly subscription pushed over monthly/yearly for faster payback period
• Paywalled immediately after onboarding questions

ASO (App Store Optimisation)

• App name format: [Name] – [descriptor] (e.g. Social Wizard – Up Your Game; CleanEats – Food Scanner)
• Subtitle carries nearly as much keyword weight as the name
• Health and fitness is the most competitive App Store niche; ranking for generic terms requires massive daily download volume
• Ratings are a ranking signal — worth engineering early
• "Inherently viral" = showing something users have never seen before; everything else follows

Creator marketing: how to build a distribution machine with no budget

• Made content himself first; validated at 5K–17K views before involving others
• Target: micro-creators with spiky view history (2–3K average, but some viral hits), not polished influencers
• Outreach script: establish credibility, frame first video as a "test" to lower price resistance, hint at long-term deal potential
• Paid $120 for the first video that drove Social Wizard's viral moment (2M views on Instagram)
• Micromanaged every script and format — creator adds style, not concept
• Minimum View Guarantee (MVG) tied to one platform only, measured at 7 days post-post; shortfall = more clips owed
• Creators care more about total payment than number of integrations — use this to negotiate more clips at the same rate
• Social Wizard creators: teens, casual, cheap, but flaky; CleanEats creators: professionals, $1,500–$5,000/video, reliable
• One CleanEats creator: $5K deal for 6 videos → ~$20K revenue returned

What makes a good reply (data from ~5M hints)

• The recipient should instinctively know how to respond — eliminate yes/no dead ends
• Embed a follow-up thread in the reply (e.g. "Have you heard In My Feelings Live? Drake's playing here" vs. "Drake's in concert tomorrow")
• Avoid open-ended statements with no clear response path

CleanEats: build, acquire, exit

• Scans packaged food barcodes; rates products on a skin score and weight score separately
• Marketed on attractiveness, not health — "become more attractive" outperforms "be healthier"
• Hard to scale: health/fitness creators already hit by Cal AI; constant mis-identification as a competitor
• Revenue hit ~$60K; tweeted intent to shut down → inbound acquisition offers → sold for low six figures within weeks
• Lesson: health/fitness apps are far more acquirable than niche consumer social apps; most consumer apps are not sellable

The immigration backstory

• Left Nigeria at ~19 with ~$100; slept on a college friend's floor
• Bounced between relatives for months after prior app (Caspid) failed
• Made homeless when an aunt called police; Maryland's squatter law meant she had to get a court order — he left voluntarily
• Moved into a second relative's home with a 5-month deadline — Social Wizard made $2K one week after he was kicked out
• Secured O-1 visa; green card in progress; total immigration spend ~$70K

Decision framework

• Every action filtered through: does this increase or decrease the likelihood of hitting my goal?
• Applies to daily behaviour (diet, city choice, who to spend time with) and business decisions (pricing ceilings, acqui-hire offers)
• Current focus: consumer infrastructure — tools to make it easier to build, market, and monetise apps