SHP AI
The groundwork for general intelligence
Data and expert judgement for expanding the frontier of AI
October 2026
What we do
We produce frontier-grade data at industrial scale.
Company records
Real work
Mail, CRM, finance, approvals and files, licensed from companies and de-identified
Expert-created
New data
Tasks, rubrics, reasoning and benchmarks written by experts to train and measure a capability
Data processing
Preparation
Prepare, label and grade data to make it ready for frontier model training
The scale of the problem
Compute at Anthropic and OpenAI, run rate late 2026
Hundreds of billions of dollars of compute at two labs alone, bottlenecked by high-quality data
How it works: The industrialization of data
Supply
Companies and experts
Companies hold years of work in mail, CRM, finance, approvals, files and code. Experts hold the judgement to grade it.
records, experts
- Sourcing
- Preparation and de-identification
- Task and environment building
- Expert grading
- Quality control
- Verification on frontier models
- Delivery in the required format
data built to spec
Customers
AI labs and hyperscalers
Post-training teams need work environments, verified tasks and benchmarks, in their own format.
We manufacture the data that trains smarter models, all the way to AGI
For labs: Three offerings
01
Company environments
A company's systems, documents and cases, rebuilt for a model to work through
- Frozen at a point in time, with working tools
- Goal-only tasks, graded on the final state
- Difficulty calibrated on the lab's models
02
Expert data
Rubrics, reasoning, reviews and preference data from PhD-level specialists across disciplines
- For SFT, RLHF and reward models
- On cases from the records, or on data the lab supplies
- Experts recruited and screened to the lab's profile
03
Benchmark tasks
Held-out tasks that test whether a model can do a job end to end
- Private eval sets the model has never seen
- Automatic graders that resist shortcuts
- Refreshed as models improve
Data you can trust, quality-controlled with full provenance, to maximize the return on compute
For companies: Earn from the records you already own
Now
Monetize your records
Companies license operating records, code and records of how staff work with AI tools to labs through us. They keep ownership, approve each buyer and scope, and we de-identify before anything leaves.
- Start with an inventory, no upload
- We match the data to lab demand and price it
- Revenue to the company, with a referral share for introducers
Next
Own your intelligence stack
The same foundry helps a company run AI on its own work, with its own models and its own data.
- Managed post-training on the company's workflows
- Managed inference
- Agent deployment, with evals on the company's own work
A long-term partner in turning a company's data into value
For experts: Help models learn, and get paid
The work
Teach the model your field
Experts prepare data, write rubrics, grade answers and build the tasks that test a model. What you mark correct is what the next models learn.
- Law, finance, medicine, engineering, science and more
- Remote, in hours you choose
Why do it
Meaningful, and it pays
AI will do part of every profession. Experts who take part decide what correct and safe look like in theirs.
- Direct input into what AI gets right
- Alongside PhDs and Rhodes Scholars
- Paid at specialist rates
Your knowledge, built into models that serve all of humanity
Why SHP AI
The best want to work with us
- Vertical global #1, strategic partners to Fortune 500 companies
- Backed by former Chairs of McKinsey and Goldman Sachs Asset Management
- A track record of attracting Ivy League PhDs and Rhodes Scholars
Quality worth the compute
- Foundry-grade volume, held to a high standard
- Quality-controlled, with provenance and expert sign-off
- Delivered on time, always on
Built and scaled before
- 100 billion records, built and run in production
- Post-training for the AI Math Olympiad world champions
- Platforms with tens of millions of daily users
The first call for data you can count on
Team
Alan
- Built a global #1 platform of thousands of Ivy League mentors and mentees
- Backed by former Chairs of McKinsey and Goldman Sachs Asset Management
- Fixed income structuring at J.P. Morgan, bespoke solutions for global institutions
- Oxford's 2012 Olympic Torchbearer, one of eight invited to meet the Queen
- Oxford, Financial Economics
Kuan
- Co-founder of AcFun, grown to tens of millions of daily users
- Brought in SoftBank and Youku as funders and strategic partners
- Built a mobile game sold to NetEase
- Investment banking at CITIC and Standard Chartered, private equity at 3i
- Imperial College London, Electronic Engineering
Shuai
- CTO and co-founder of Mowgli (acquired by Hopper) and Jinko (Y Combinator)
- Ran travel data infrastructure at 100‑billion‑record scale
- Early member of Numina, the AI Math Olympiad world champions, and post-training with Mistral
- École Polytechnique and Télécom ParisTech, Machine Learning
Founders who have built platforms, data systems and models at scale
SHP AI