Skip to content

We put AI to work inside the business.

Not tools to evaluate. Not pilots that stall. We design, build and operate the AI systems that carry real work every day — end to end, in production, measured against your own numbers.

Hong Kong based. Working with teams across Asia and the United States.

How we work

Three steps, and we own all of them.

  1. 01

    Find the work worth doing

    We start in your operation, not in a model. Which processes are slow, expensive, or capped by headcount? Where does judgment genuinely need a person? You get back a short list ranked by what it is worth — including the parts where we think AI is the wrong answer.

  2. 02

    Build it to production standard

    We build on whichever models fit the task — Claude, GPT, open-weight — behind an evaluation harness that proves the output holds before anyone depends on it. Delivered as a managed service, or deployed entirely inside your own environment when the data cannot leave.

  3. 03

    Run it, and keep it honest

    An AI system nobody watches quietly drifts. We operate what we build: monitoring, evaluation, model upgrades, and a plain account of what it produced. When a number stops moving, it is our problem to fix.

Why this works

The things clients tell us made the difference.

Model-agnostic by design

We are not reselling one vendor's model. We choose per task across frontier and open-weight models, and we swap when something better ships — without rebuilding your system around it.

Your data stays yours

Managed service when speed matters most, fully private deployment when the data cannot leave your walls. Your content is never used to train anyone's model.

Operators, not advisors

Two founders with two decades each of shipping and running production systems — in marketing technology and in regulated financial services. We write the code and we carry the pager.

Proof before scale

Every engagement puts one workflow into production early, where you can judge it against real work. Scope grows from evidence, not from a roadmap slide.

Field log

What we've shipped.

  • AGENCY

    US short-video marketing agency

    Engaged 2024

    1,000+ creators and influencers producing for TikTok advertisers. We automated the brief → script → asset → edit pipeline, taking per-video cost and turnaround from weeks down to hours.

  • MARKETPLACE

    US digital marketing talent platform

    Engaged 2024

    5,000+ brand clients and 50,000+ vetted marketing freelancers. AI workflows for brief generation, matching, and consistency review across a very high volume of client work.

Client names withheld by preference. We'll walk you through the specifics on a call.

Who's behind this

Two founders. Forty years of production systems.

Small on purpose. The people who scope your work are the people who build and run it.

Meng Ge

Founder & CEO

Generative-AI architect and serial founder. Meng has built and run production software across HR and finance SaaS, recruiting platforms, mystery-shopper and RPO/BPO operations systems, fund trading and robo-advisory platforms, and consumer search and social apps. He leads client work and delivery.

Yiming Zeng

Co-founder

More than twenty years in financial services technology — as programmer, architect and director — including Morningstar and Haitong Securities. Yiming brings the discipline regulated environments demand: uptime, auditability, data control, and systems that are still maintainable years later.

More about us

Start with a conversation, not a proposal.

Thirty minutes. Tell us where the work piles up, and we'll tell you honestly whether AI is the answer — and if it isn't, we'll say so.

Book an introduction