Orca21 Labs is building a self-improving system that keeps every company on the best AI, and gets better at it with every company it serves.
3 signed pilots
All three signed in our first week of selling Orcapo.
10+ calls
Booked with more companies that want to try it.
Google and MIT
Our CTO is first author on AI research with both.
A sales team, a support desk and a research lab each do their best work when their tools are set up around their own tasks, rules and goals. Setting that up takes an expert, and AI gets better every few months, so a good setup goes out of date fast. Companies have tried to solve this in stages. The diagram below steps through them, and the red line under each stage says what it still misses. None of the first three tests new AI on the company’s own work and learns from results which changes work.
Stage 1 · ChatGPT, Claude, Copilot
AI seats for everyone
The company pays for an AI login for each person. Each person works out on their own what to use it for.
Missing: Nobody picks the tasks, trains people or checks the results, so most people go back to the old way.
Stage 2 · AI consultancies, freelancers
A one-time setup
An expert maps the work, picks tools and builds a few automations. Then the project ends.
Missing: Better AI ships every few months. Nothing re-tests the setup, so it is out of date within the year.
Stage 3 · Online courses, vendor academies
An AI course for everyone
Everyone takes the same course on how to use AI, with general tips and examples.
Missing: The course is not about the team’s own tasks, so little changes at work.
Stage 4 · What Orca21 Labs is building
Recursively self-improving systems
Each company gets its own setup and its own strategy model. Every month, new AI is tested on the company’s past work, and an expert checks each change before the owner approves it. Across all companies, Pod will learn which upgrades win.
The result is a gap that keeps growing. 95% of company AI pilots show no measurable effect on profit, and only 36% of workers feel properly trained to use AI. Meanwhile AI agents went from finishing 12% of real computer tasks to 66% in two years.
We want an AI system that gets better at running each company’s AI with every company it serves, and that also gets better at the way it improves. That is RSI?Recursive self-improvement. A system that gets better at its job, and also gets better at the way it improves itself., applied to how companies work.
Human experts define what good work looks like. The system will learn from their corrections, and over time it will need them less.
It works as three loops, one inside the other.
The middle loop is Pod?The part of our system we are building to learn across companies. It pools test results from 5 or more companies at a time, with no names and no documents, and finds which AI wins for which job.. Every test we run shows which AI wins for which job. Pod pools those results across companies, so each new company starts with what already worked for others.
The same engine already runs in tutoring, and now in work.
Every step runs on the same engine, so what it learns in one place carries over to the others. We are starting with Orcapo.
The road to RSI?Recursive self-improvement. A system that gets better at its job, and also gets better at the way it improves itself. is long, so we are building toward it step by step. The first step is Corporate 0.1, our system inside Orcapo, our product for companies. It builds a harness?Everything we build around the AI model for one company: its map of jobs, its tools, its rules, the checks on the AI's work, and an onboarding guide for each person. for each company and tests every change on that company’s own past work. Today an expert approves every change. Over time, the steps we repeat move into software, so each expert can look after more companies.
Here is one round of it, start to finish.
Our system inside Orcapo, for companies of 10 to 199 people. Every month it tests new AI on the company’s own past work. An expert approves, corrects or rejects every change, and the owner says yes before anything goes live.
The film follows one round at Tidewell, a fictional software company. The screens are product designs with example data.
Three companies signed pilots in the first week we sold Orcapo, and we have more than 10 calls booked with others. Billing starts at each kickoff in November.
After Corporate 0.1, we will build Pod and then Corporate 1. Each one starts once the step before it meets its target.
The part of our system that will learn from all companies at once. With each company’s consent, Pod will compare test results across companies, find the upgrades that win for each job, and add them to the starting setup every new company gets. It shows a result only when 5 or more companies are in it.
Our goal: a company that improves itself. A company will enter its goals and rules, and Corporate 1 will choose, test and roll out its AI tools and training, then keep re-testing them as new AI ships.
We first built this loop for LessonOrca, our tutoring company. After every session, a strategy model proposed a change to the student’s plan, and a reviewer approved it. Our first students gained about 150+ points on the SAT. LessonOrca now runs on its own, and we have moved our focus to Orcapo.
Co-founder and CTO
Self-supervised learning, foundation models and learner modeling
First-author paper with Google and MIT (IEEE Access, 2026): a self-supervised foundation model for spectroscopy that trains with 60–90% less labeled data
Technical lead, U.S. Treasury EyeNote: built its currency recognition model (99.9% precision) for 100,000+ blind and low-vision Americans, starting in high school
Inventor on a provisional patent in impedance spectroscopy: built Chronus Health’s best 3-part white blood cell classifier, in use for over a year
Architect of the LessonOrca Student Profile: turns hundreds of observations about a student into one context record
Built ZemoMemo: models each learner’s forgetting curve (FSRS-6) and schedules recall for 1,000+ learners
Co-founder and CEO
AI infrastructure, education businesses and go-to-market
AI/ML consultant at Databricks: infrastructure research with field engineering and product: lakehouse architectures, open table format benchmarks, enterprise AI strategy
Founder and CEO of WEInstruct, an SAT and ACT tutoring company: personalized plans for 30+ students, 5 tutors, $10,000+ revenue
Founder of Tin2Totes, a profitable recycling operation: 50+ collection sites across Washington and Oregon, recurring monthly revenue
Founder of The Bag Project: raised $10,000+ and distributed 10,000+ units across four states and Pakistan
Washington DECA State Vice President: elected to lead 15,000+ members across 15 regions; International Finalist and State Champion, National Economics Challenge
Berkeley SkyDeck, Pad-13 (Batch 22) · UC LAUNCH accelerator, Fall 2026
We are raising a $1M pre-seed round on a post-money SAFE.
Email us at [email protected].