The AI Mastery Platform — for those who'll define the next era

AI is growing fast. So should you.

For thousands of years, training was about learning to finish things. But finishing — implementation — is exactly what AI now does, and fast. Your value moves to how far you can amplify yourself through AI: real fluency with it, solving complex problems alongside it, and aiming it precisely at your own domain. And putting AI to work where it truly counts — manufacturing, medicine, agriculture, business, insurance, security — takes far more than ChatGPT or Claude Code. It takes mastery, and the platform to build it.

2,000+engineers trained
Silicon Valleyresearch & eng team
San MateoCalifornia HQ
Worldwidelearners & customers

Why Parallight is different

The Agent era changed what it means to learn. We built for each shift.

Four shifts moved the ground under traditional training — and the four things we made in answer.

GIF · PILLAR 01Agentist — the frontier, streaming in daily

01 · Continuous, for life

Learning never stops now.

As AI reshapes every industry, staying current is no longer something you finish in school or at onboarding. It's a lifelong requirement — the frontier moves every week.

→ Agentist. A continuous source of frontier information — the day's dispatches, papers and model releases, read so you never fall behind.

Read Agentist →
GIF · PILLAR 02A live sandbox — a real terminal, running

02 · Learn by doing, for real

Understanding isn't the skill. Doing is.

The value of the new intelligence lives at the application end. You don't learn it by taking in concepts — you learn it by building, running, and practising on real problems.

→ An online sandbox. A real runtime environment where every lesson is hands-on: a real terminal, a real agent, real output — nothing simulated.

Try the sandbox →
GIF · PILLAR 03A private Mentor Agent — slides, code, feedback

03 · Personal, at any depth

One curriculum can't fit everyone.

Everyone's background and domain differ — and this technology's depth, breadth and iteration speed outpace any human-taught, one-size course. A private mentor for each person is the only thing that keeps up.

→ A Mentor Agent. Fully private and personalized, able to teach complex tasks: slides, interactive walk-throughs, code analysis, live code runs, and immediate feedback.

Meet the Mentor →
GIF · PILLAR 04A cloud IDE — long-horizon runs, searching for solutions

04 · Find, don't just build

The work moved from building to finding.

Code agents can now implement almost anything. So the human's job shifts: define the right evaluation metrics, let the AI search for solutions, and verify what comes back. Finding solutions and research are eclipsing hand-implementation.

→ A cloud IDE. Built for ultra-long-horizon tasks: intelligent long-running with dynamic CPU/GPU auto-management — your platform entry for solution-finding.

Open the IDE →

How you engage

Three ways in.

SELF-SERVE

The Lab

Hands-on labs where you command an agent to build real things — research agents, harnesses, eval suites — and profile what it ships.

Enter the Lab →

GUIDED

The Cohort

A live training program: structured, personalized, with real projects and feedback. For teams and individuals who want to go deep, fast.

See the cohort →

CLOUD

Online IDE

A browser IDE with a real terminal, a real agent, and automatic CPU/GPU — nothing to install. Long-horizon runs that finish the job.

Open the IDE →

Who's behind it

Built by a Silicon Valley team.

A research-and-engineering team from Silicon Valley, headquartered in San Mateo, California. Over 2,000 engineers have trained at Parallight and in the founder's earlier programs, all over the world.

2,000+engineers trained
San MateoSilicon Valley HQ
Worldwide13 countries · 4 continents

The Founder

Marvin Gao

Founder & CEO · KeploreAI / Parallight

AI scientist turned serial founder. Built production AI at IBM, Ant Financial, and Alibaba, and previously founded an AI-education platform acquired by a unicorn. Now building Parallight for the Agent era.

ex-IBM AI ScientistAnt · AlibabaFounded & sold an ed-tech platformPeking Univ · JHU

Where we sit — and how far we reach

One point on the map. The whole world through it.

Parallight — parallel light. A small point in the Bay Area, where the world's frontier converges and goes back out to learners everywhere — ringed by the companies building AI.

2,000+engineers trained
13countries · 4 continents
14AI giants next door
1point · the Bay Area

Our publication

Agentist — a daily on the agent frontier.

An Economist-style edition, written by our agents and edited by us: what an Agentist is, plus the day's dispatches, papers, model releases and editorials — every item runnable in a live sandbox.

Ability is trained, not granted.

Start in a live lab — a real terminal, a real agent, nothing to install.