Skip to content
AI Dev Cases
Cal AI

Built in High School at 17 — How Cal AI’s “Just Snap Your Food” Calorie Tracker Hit ~$50M ARR in 18 Months and Sold to MyFitnessPal

A photo calorie tracker built by Zach Yadegari and a teenage founding team: snap your meal and AI estimates the calories. Bootstrapped to ~$50M ARR and 15M downloads in 18 months, it was acquired by MyFitnessPal in 2025.

Published: Jun 24, 2026Figures last checked: Aug 13, 20262 min readPrimary-source verified · 2
Monthly (est.)
$4.2M/mo
Time to grow
18 months
Users
15.0M
Launched
2024
Zach YadegariZach Yadegari@zach_yadegariBuilt in High School at 17 — How Cal AI’s “Just Snap Your Food” Calorie Tracker Hit ~$50M ARR in 18 Months and Sold to MyFitnessPal

Key takeaways

  • Erase a universal chore in one action (no searching or weighing — ‘just snap it’)
  • Don’t train a giant model; translate existing powerful AI (GPT) into the product experience
  • Before outsourcing, become the TikTok creator yourself to teach the algorithm your audience

The pain point, and how they found it

Calorie logging is a near-universal point of failure for anyone trying to lose weight: searching every food and entering portions is tedious, so people quit within days. Cal AI attacked that logging friction head-on — just take one photo of your meal. The ‘entry is a chore’ complaint baked into incumbents like MyFitnessPal was itself the market entry point.

Background & product

Cal AI estimates calories and macros (protein, fat, carbs) from a photo of your meal. It also supports barcodes and manual entry, but the core is the ‘just snap it’ experience that removes tedious searching and weighing.

It was built by a team of teens and early-twenties founders led by Zach Yadegari. Zach (CEO) taught himself to code at 7 and started Cal AI at 17 while still in high school. He was joined by Henry Langmack (CTO), whom he’d met at coding camp, plus Blake Anderson — who he met on X and who had a track record of consumer AI apps like RizzGPT and Umax — and Jake Castillo. They launched in May 2024.

Technically, they used OpenAI’s GPT models for the photo-to-calorie estimation rather than training a giant model from scratch — a strategy of ‘translating existing powerful AI into a product experience.’ Frontend in Swift, backend in Node.js/Python, Firebase for infra, and Superwall for paywall optimization: a stack an indie team can reproduce. Pricing was $10/month or $30/year, prioritizing adoption over per-user revenue.

Eighteen months after launch, having reached 15M downloads and ~$50M ARR, Cal AI was acquired by MyFitnessPal — the longtime synonym for calorie tracking. The incumbent known for tedious entry absorbed the teenage app that erased exactly that pain.

Pricing

Free to download with in-app purchases. A subscription is required to see AI photo-analysis results (listed in-app purchases run $2.99–$29.99, plus a $0.99 “Streak Restore”). The App Store listing does not spell out the billing period for each item.

App screens

  • Home dashboard showing “1250 calories left” alongside rings for remaining protein, carbs and fats, with recently logged meals (a fattoush salad at 153 calories, etc.) listed below with photos
  • Food-scanning camera screen: a recognition frame overlays a plate of pancakes, with a “Scan food” mode selector and shutter button at the bottom
  • AI analysis result for the photographed meal: callouts over the image break it down by ingredient (blueberries 8, pancakes 595, syrup 12), with totals below — 615 calories, 93 g carbs, 11 g protein, 21 g fat and a health score of 7/10
  • Progress screen: a 90-day weight line chart at the top (currently 67 kg, 80% of goal achieved) and, below it, a daily stacked bar chart of the week’s calorie intake colour-coded by protein, carbs and fat

Images are served directly by the App Store / the product's own site (not rehosted here) — Cal AI

From the founder (primary source)

Cal AI growth channels and tech stack

How it was built

  1. Erase a universal chore in one action — no searching, no weighing, just snap the meal
  2. Don't train a giant model in-house; implement calorie estimation on OpenAI's GPT models (translate existing AI into UX rather than building it)
  3. Use a stack a solo developer can reproduce: Swift on the front end, Node.js/Python on the back end, Firebase for infrastructure, Superwall for paywall optimization
  4. Price for adoption over ARPU at $10/month or $30/year
  5. Continuously A/B onboarding and the paywall in Superwall, settling by the numbers which screens in which order lead to a purchase
What AI did
Calorie estimation runs on OpenAI's GPT models as-is, with no in-house model training — the edge is translating strong existing AI into a frictionless experience, not building frontier AI

How it got users

  1. Before hiring any influencer, become the TikTok creator yourself — engage only with health and fitness content so the algorithm learns which audience to serve
  2. Post your own usage videos and manufacture the first viral hit by hand (100K downloads from organic alone, giving you demand signal and the winning angle at zero CAC)
  3. Turn the angle you found into a brief and have micro-influencers — not big names — mass-produce native videos that don't look like ads ($2M/month within months)
  4. Only after demand and the winning angle are confirmed, go all-in on performance ads across Facebook, TikTok and Instagram, using the creative that already won in stages 1-2 ($1M+/month spend against $5.7M/month revenue)
  5. For the first six months, cover App Store payout delays out of the founders' own pockets and keep the cash cycle turning
First users
Organic traffic from the founder's own TikTok account (100K downloads)
Early ad spend
$0 at first (the founder's own organic reach), then performance ads once demand was proven — over $1M/month spend against $5.7M/month revenue as of January 2026

The repeatable playbook

  1. 1Erase a universal chore in one action (no searching or weighing — ‘just snap it’)
  2. 2Don’t train a giant model; translate existing powerful AI (GPT) into the product experience
  3. 3Before outsourcing, become the TikTok creator yourself to teach the algorithm your audience
  4. 4Commission micro-influencers to mass-produce native, ‘doesn’t-look-like-an-ad’ videos (→ $2M/mo)
  5. 5Validate demand organically, then go all-in on performance ads (build payback first, then spend)
  6. 6Price for adoption over ARPU ($10/mo, $30/yr) and continuously A/B the paywall with Superwall

Channels they actually used

The hard parts

Behind the headline numbers, for the first six months the founders fronted operating and marketing costs out of pocket to bridge the app stores’ delayed payouts. Even as teenagers, the foundation was unglamorous cash-flow management — keeping it running before the cash ran out.

Updates since publication

  1. Aug 13, 2026

    The MyFitnessPal acquisition closed in December 2025 and was announced on March 2, 2026. At announcement, trailing-12-month revenue was ~$40M with ~$50M projected for 2026. Cal AI continues as a standalone product, and a seven-person team including co-founder Zach Yadegari was retained.

Deep dive

【Deep dive】We break down the ‘three-stage acquisition rocket’ that took a near-bootstrapped teenage team to ~$50M ARR in 18 months — stage by stage. What matters is less *what* they did than the *order* they did it in.

Stage 1: The founder becomes the TikTok ‘creator’ first (→ 100K downloads). Before paying any influencer, Zach grew his own TikTok account. The method was unglamorous: engage *only* with health/fitness content so the feed (and the algorithm) learns ‘distribute to this audience,’ then post app-in-use videos himself and spark the first virality by hand. This worked because (1) it reads demand at zero CAC, (2) he personally learned which angles land, and (3) he could hand that winning formula straight to the next stage’s creators. Starting with outsourcing means outsourcing that learning too — and losing repeatability. Organic alone got them to 100K downloads.

Stage 2: Micro-influencers making ‘doesn’t-look-like-an-ad’ videos (→ $2M/month). Next, fan out — not to mega-influencers but to high-engagement micro creators, commissioned to post in native style (their normal tone, not an obvious ad). Because the brief encodes the winning angles found in Stage 1, the hit rate is high. Killing the ad-feel works because TikTok users skip ads instantly but trust a creator’s genuine pick. This format scaled them to $2M/month within months of launch.

Stage 3: Validate demand, *then* go all-in on performance ads (→ $5.7M/month). Only after organic and influencer proved ‘there is demand, and we know what lands’ did they pour into FB/TikTok/Instagram performance campaigns — using the already-validated videos as ad creative, a massive head start over testing from scratch. By January 2026 they spent $1M+/month against $5.7M monthly revenue. Reverse the order — ads before validation — and you just burn cash on creative that doesn’t land.

Pricing & paywall: take adoption over ARPU. Pricing was an aggressive $10/mo, $30/yr. Raising it lifts per-user revenue, but they chose adoption: a photo calorie tracker spreads well via word of mouth and social, so user count itself fuels the next wave of acquisition (UGC, reviews, referrals). They also A/B-tested onboarding and the paywall continuously with Superwall, dialing in *which screens in which order* convert. A low price still works if you raise conversion and retention operationally.

Tech: ‘translate,’ don’t ‘build.’ They didn’t train a giant model for calorie estimation — they used OpenAI’s GPT models. The lesson for indie builders is clear: the edge isn’t building frontier AI yourself, it’s translating existing powerful AI into a frictionless product experience. Cal AI’s moat was the snap-only UX and the acquisition playbook, not the model.

The overlooked foundation: unglamorous cash flow. Behind the headline numbers, for the first six months the founders fronted operating and marketing costs out of pocket to bridge the app stores’ delayed payouts. Even as teenagers, grounded cash management — keep it running before the cash runs out — underpinned everything.

Cross-case “growth playbook” report (coming soon)

We're building a paid report that aggregates every case in this database: which acquisition tactics worked, in which categories, and how well — insights you can't see from a single story. Get notified first when it launches.

Related cases