Ben Or-Chen

1.5 yrs ago, a VC offered $3M to drop consumer apps and switch to anything b2b because “conSUmeR is HArD” but we said no since then: - product 1 hit 1M+ users and profitable in 9 months - product 2 hit 1M+ users and profitable in 9 weeks - this week reminded us that we made the right call and we are literally unstoppable we’re profitable we’re growing exponentially we’re coming after every market and we’re hiring! more than halfway through to $20M ARR by eoy ;) looking for product engineers, ai engineers, and designers apply on our website, or email talent [at] oleve [dot] co

Come hear me yap at the @aiDotEngineer World’s Fair (June 3–5, 📍SF) giving 2 talks: #1 June 4, 11:35 AM: “The New Lean Startup”. sharing our playbook on how we skyrocketed Oleve from $1M to $6M ARR and top in the charts with just 4 people, our two-track eng system, philosophy for scaling agents, and scrappy infra hacks that drive our lean growth #TinyTeams #2 June 5, 2:20 PM: “Testing the Un-Testable: Monitoring AI Products in the Wild”. w/ @benhylak from (@raindrop_ai) where I’ll share TRELLIS, my framework for scaling reliable consumer AI UX powering 5M+ users, designed around our virality engine, 500M+ views on socials. We’ll share how good UX isn’t a vibe, it’s engineered #AIArchitects come thru to spot the differences between me and this cursed ai-generated version

In the last 2 years I’ve scaled viral agentic AI experiences to 5m+ users globally, $6M ARR this is my playbook for designing reliable AI experiences around virality, so UX doesn’t suck, designed around our viral growth engine: 500m+ views across socials TRELLIS: Targeted Refinement of Emergent LLM Intelligence through Structured segmentation it is the framework I use to design, orchestrate, and iterate on retentive consumer AI experiences philosophy: > magic must be engineered, repeatable, testable, and attributable, not accidental > systems must be improvable through isolated, blameable workflows > in a world of shitty AI products, you must win both the ‘novelty tests’ and the retentive flows principles: > discretization - translate an infinite output space into distinct, mutually exclusive buckets > prioritization - score and rank buckets using impact-driven metrics > recursive refinement - continuously repeat the process within prioritized buckets > constrained expressiveness - create semi-deterministic workflows, broad enough to be useful and narrow enough to be reliable 6-step framework for success: 1. launch open-ended mvp agent experience guided by initial priors 2. look at your data and cluster it into interactions by user intents 3. turn these intents into semi-deterministic workflows 4. use a well-defined scoring system to prioritize workflows that actually move the bottom line. I recommend: volume x negative sentiment x achievable delta x strategic relevance 5. analyze the high-priority workflows for sub-intents and failure patterns 6. recursively apply the TRELLIS method within priority sub-workflows presenting TRELLIS on Wednesday 28th as part of @jxnlco 's course I'll cover how we're using @raindrop_ai to power this framework with @benhylak read more about the framework on the oleve website