How Oleve Built a $6M ARR AI Empire w/ 4 employees from a Homework App (Quizard AI/Unstuck AI)

Sid Bendre and his team at Oleve have cracked the code for building viral, profitable AI consumer apps. With just six full-time employees, they’ve grown to over $6 million in annual recurring revenue and 5 million users, all while staying lean, culturally fluent, and systemically engineered.

This is the story of how they went from a dorm-room project to a consumer AI studio challenging TikTok and Google.

From a College Side Project to a TikTok Explosion

In early 2023, students at the University of Rochester launched Quizard, a math homework-helper app that let users scan problems and receive AI-driven tutoring. The backend was stitched together using Python, Flask, Notion, and Excel. They routed API calls through 10 different OpenAI keys, relying on GPT-powered completions and Codex.

One TikTok video with the hook "POV: ChatGPT and Photomath had a baby" exploded to 2.3 million views overnight, onboarding 10,000 users in under 30 hours.

Within weeks, they added a $9.99/week subscription, proving Gen-Z’s willingness to pay for useful AI apps.

Engineering Feedback Loops Instead of Just Code

Post-viral traction, the team joined the Neo accelerator and relocated to NYC. From there, they focused on turning their momentum into repeatable processes:

• Built modular prompt-routing workflows

• Introduced LaunchDarkly feature flags to distribute OpenAI usage

• Created Excel-driven eval frameworks for rapid iteration

• Transitioned from Codex to GPT-3.5 after Codex was shut down

They initially received $2,000 in Azure credits, later expanding to over $300,000, using that to scale while minimizing costs. One clever hack: using LaunchDarkly flags as a pseudo-load balancer across multiple Azure OpenAI endpoints.

Sticky Notes and 8 Million Users in Nine Weeks

In August 2023, Oleve launched Unstuck AI, a study assistant built for virality. Their campaign used handwritten sticky notes, trending TikTok audios, and authentic video editing that mimicked top creators like MrBeast.

• Over 250 million TikTok views in under 30 days

• Reached #4 in the App Store education charts

• Amassed 8 million users in just 9 weeks

Distribution extended across Instagram Reels, Discord communities, and automated TikTok ad tests. The entire GTM was backed by systems—automated agents scraped trending audio and generated viral concept tests.

From Viral Wins to a Repeatable Product Studio

Oleve soon transitioned from app builder to platform operator. They reorganized into three units:

• Product Engineers (Harvesters): Run apps like founders—UX, monetization, GTM.

• AI Engineers (Cultivators): Build shared infra: prompt routing, referral systems, eval pipelines.

• Agents Team: Automates internal operations—TikTok trend detection, content scripts, Discord sentiment scraping.

This modular design allows fast iteration across verticals, whether it’s study tools, wellness, or stealth projects.

TRELLIS: Turning Open-Ended AI Into Predictable UX

One major challenge of building consumer AI is consistency. Oleve’s answer: TRELLIS (Targeted Refinement of Emergent LLM Intelligence through Structured Segmentation).

It works like this:

• Let users interact with open-ended agents

• Cluster their queries into high-volume intents

• Build deterministic workflows for each intent bucket

• Measure performance via sentiment, error rates, and UX feedback

• Refine recursively to stabilize results without losing the magic

TRELLIS allows their apps to feel both magical and reliable. It makes debugging fast, A/B testing precise, and performance predictable.

Why They Think Like Duolingo and Coca-Cola, Not OpenAI

Oleve doesn’t sell general-purpose AI. They build culturally sticky brands. Their true competitors aren’t OpenAI—it’s brands like Photomath (Google), Gauth (TikTok), and Duolingo.

By studying TikTok 24/7 and reverse-engineering viral content, they:

• Design products with shareability baked in

• Use sticky notes and Gen-Z aesthetics as onboarding UX

• Embrace churn, betting on re-installs via cultural recall

As Sid puts it, they benchmark against Coca-Cola: “It’s okay if someone uninstalls. We want them to remember the name.”

$500K Monthly Revenue, Profitable in 9 Months

Today, Oleve generates over $500,000/month in revenue, hitting $6M ARR with just a handful of employees. Their profit margins range from 20–30%, a rare feat among AI startups.

Their monetization strategy leans on:

• Mobile subscriptions tested with Superwall

• Pricing strategies optimized for Gen-Z preferences (weekly plans outperform annual)

• In-app referrals, native virality, and minimal paid acquisition

They’ve raised a small angel round post-Neo, including investments from Cal Henderson (Slack), Mary Jang (Tinder), and Russell Kaplan (Cognition Labs).

Agents That Replace Entire Teams (Almost)

Oleve’s most ambitious tech isn’t user-facing—it’s internal agents:

• Growth agents scrape TikTok trends, map viral formulas

• Research agents identify verticals with high pricing potential

• Creator agents source influencers and generate UGC-style campaigns

The future they envision? A "launch engine" where agents do 80% of the work, and humans only focus on taste, strategy, and creative iteration.

Competing With TikTok, Google, and Zuoyebang—And Winning

Despite competition from TikTok’s Gauth, Google’s Photomath, and Zuoyebang’s Question.AI, Oleve remains competitive through faster execution, culturally aligned branding, and tight, measurable feedback loops.

They don’t fear saturation. In fact, they’re planning to expand into new verticals beyond education, using the same go-to-market playbooks that made Unstuck a success.

The Pixar of Consumer AI Apps

Oleve sees itself not as a company with one hit app, but as a repeatable studio. Their model is closer to Pixar or Marvel than to most tech startups. They ship quickly, measure rigorously, and adjust based on real-time feedback.

What powers that engine:

• Taste in brand and UX

• Systems for shipping quickly

• Culture of experimentation without bloat

In a world where most AI startups are demos, Oleve is building a machine that scales magic.

Hat Tip to Latent Space and Nathan Latka for the insights.

Top Tweets of the day

1/

AI-generated Podcasts have arrived. I bet this newsletter converted into a 5-minute podcast would do well.

Reading takes effort. Podcasts doesn't. It is passive. You can do reading passively but it takes a lot of energy.

Listen to a 40-minute podcast and it feels like nothing. Do a 40-minute read and you'll see your brain becoming more active.

2/

This is why Peter Thiel funds 19 and 20-year olds and Elon hires youngsters to do the work of their life. The older you are, the more obligations you have.

3/

Good hack if you want to find tweets from particular people via APIs. Or you can use Grok 4 to schedule tasks to get them daily that you get for even free accounts.

Rabbit Holes

1. User Embeddings: How TikTok Knows You Better Than You Do by Lossfunk

2. 10 proven ways to improve email open rate by Matthew McGarry

3. Marketing from Mad Men to AI by Deloitte

What’d ya think of today’s newsletter? Hit ‘reply’ and let me know.

Do me a favor and share it in your company's Slack #marketing channel.

First time? Subscribe.

Follow me on X.

More Startup Spells 🪄

1. How Kilo Code's Naming Hack Hijacked Search Results From Popular Competitors (LINK)

2. How Family and Group Plans Improve SaaS Retention and Reduce Churn (LINK)

3. Trikko's Faceless Marketing Playbook (LINK)

4. How South Korean Government Uses Incentives Like Cash and Apps to Solve Its Aging and Health Crisis (LINK)