Explore AI is a research and product company building adaptive agents, evidence-led systems, efficient models, and interactive worlds. Discovery makes financial thesis review inspectable. OnlineNIW turns complex careers into source-linked evidence records for counsel review. Photon47 turns consequence into a playable world.
Detailed interpretive reconstruction. Landmark meshes are procedurally modeled from public architectural references, but are not survey-accurate digital twins. Earth geometry uses public-domain Natural Earth coastlines. The named cities are narrative waypoints; they do not represent company offices or a corporate chronology. Product panels do not depict live data.
Shanghai · Irvine · San Jose · San Francisco
Ideas travel. Evidence compounds.
Scroll through filled, shaded procedural models of Shanghai, Irvine, San Jose, and San Francisco—waterfronts, terrain, civic space, bridge structure, and recognizable buildings—then pull back to a Pacific-centered Earth built from real Natural Earth geography. On supported desktops, pointer movement shifts the view; every chapter remains readable and operable without motion.
01 Shanghai · Observe
Read the system before changing it.
The camera enters beside a river and moves between a dense skyline of Pearl-inspired silhouettes. Interacting scales introduce the first discipline of useful intelligence: notice structure, motion, and constraint before reaching for an answer.
31.2304° N · 121.4737° E
Landscape
Huangpu · Bund · Lujiazui
Geometry
Lofts · lattices · glass shells
Reference scale
632 m → 421 m
Select a building to inspect its procedural construction.
02 Irvine · Frame
Give difficult questions room.
The street-level route opens into landscape and radial campus geometry, shifting the view from density to deliberate inquiry: define the question, preserve uncertainty, and make the route to an answer inspectable.
33.6846° N · 117.8265° W
Landscape
Park bowl · palms · low hills
Geometry
Canopy · rigging · wheel truss
Urban grain
Planned · green · low-rise
Select a landmark to inspect its procedural construction.
03 San Jose · Build
Turn research into working systems.
The view drops between civic curves, compact towers, and a precise street grid for the engineering step: reduce friction, respect constraints, and make capability dependable enough to use.
37.3382° N · 121.8863° W
Landscape
Guadalupe · plaza · foothills
Geometry
Rotunda · dome ribs · truss
Civic scale
Arcade → 18-story tower
Select a landmark to inspect its procedural construction.
04 San Francisco · Engage
Bring ideas to the edge of consequence.
The last street passage climbs through hills toward bridge and vertical skyline forms—a meeting point for product craft, capital, and public responsibility. The question becomes not only what can work, but what should earn trust.
37.7749° N · 122.4194° W
Landscape
Bay · piers · peninsula hills
Geometry
Catenary · deck truss · taper
Vertical range
235 ft → 1,070 ft
Select a landmark to inspect its procedural construction.
05 Discovery · Evidence to practice
Make every market decision inspectable.
Discovery connects source-linked evidence, an interactive candle-and-volume lab, an editable deterministic strategy graph, bounded search, foreground signal receipts, and future-blind trader training. The immediate price path is a labeled offline fixture; protected market data and private services remain deployment-gated.
Inspectable market workflow
Evidence
Chart
Graph
Practice
Evidence → test → practice. The diagram is conceptual—not market data or a performance claim.
06 OnlineNIW · Record
Make a complex career reviewable.
OnlineNIW helps researchers, technical founders, and senior professionals organize a source-linked claim-to-exhibit record with contradiction checks and human review gates. Operated by Explore AI Inc., it is not a law firm and does not provide legal advice or guaranteed outcomes.
Synthetic interface diagram. Client, recommender, and counsel review remain distinct.
07 Photon47 · Act
Make consequence visible through play.
Photon47 turns a connected biological universe into a playable space where regions, roles, movement, combat, and feedback make consequence visible. It is an interactive creative work, not a scientific simulation.
Each region changes the rules, roles, and consequences.
08 Earth · Learn
Pull back. See one spherical world.
After the four street-level passages, the camera rises beyond the skyline and resolves the journey on a true spherical map. Natural Earth coastlines, country boundaries, lakes, and latitude/longitude-positioned city markers replace the old invented connection network; our agents likewise gather evidence, test revisable hypotheses, and change plans when outcomes break expectations.
Street to orbit · four exact waypoints
ShanghaiObserve
IrvineFrame
San JoseBuild
San FranciscoEngage
Annotated waypoints become process: observe, model, act, learn, repeat.
Shanghai · Irvine · San Jose · San Francisco · Discovery · OnlineNIW · Photon47 · Earth
02 Research landscape
Ideas become useful where fields connect.
Navigate the questions, systems, evidence, and worlds that shape our work. Every node opens a concise case-study page; hover or focus to preview its role in the larger field.
Conceptual research map—not measured data, a product roadmap, or a claim that every path is solved.
03 How we think
Learning is a trajectory, not a snapshot.
We design for the whole loop: perception, world modeling, purposeful action, and evidence-led adaptation.
Explore AIObserveModelActLearn
01 · Observe
Attend to the world.
Ground decisions in multimodal evidence, interaction history, and the parts of an environment that can actually change an outcome.
02 · Model
Make uncertainty legible.
Build compact representations of state, cause, and possibility—then expose where those representations are incomplete.
03 · Act
Choose with purpose.
Translate reasoning into measurable interventions. Good plans should survive contact with dynamic environments, limited resources, and other agents.
04 · Learn
Let evidence change the system.
Use outcomes—not confidence alone—to update memory, policy, and the questions the system asks next.
04 Evidence
Measure progress without mistaking a metric for the mission.
Benchmarks can reveal capabilities and failure modes. We report them with context and keep the larger question open.
ARC‑AGI‑3 · Kaggle public score
25.7+
At our October 5 snapshot, our ARC‑AGI‑3 entry ranked 618th of 3,800 teams on Kaggle’s public leaderboard.
That 25.7+ public score is provisional while the competition runs. It is not a final private score, final placement, or awarded medal—and it is not evidence that a system has achieved artificial general intelligence.
What the work examines
Exploration, world modeling, goal acquisition, and planning in unfamiliar environments.
These environments help us study how an agent gathers information before acting, how it recognizes useful state, and how quickly it can adapt when a first strategy fails.
Our products make reasoning, evidence, and consequence visible—so people can examine a decision process or learn a world through action.
Market research + trader training · public workflow available
Discovery
Move from source-linked evidence to chart inspection, an editable strategy graph, bounded parameter search, foreground signal receipts, and future-blind decision practice. The immediate public lab is deterministic and clearly labeled; the 3:32 walkthrough shows the complete workflow before you open it yourself.
Available now Offline market lab, source-linked NVIDIA case, editable deterministic DAG, bounded browser search, foreground signals, future-blind trainer, and product tour. Deployment-gated Authenticated Alpaca U.S. history, server trainer/history, private model briefs, and paid enrollment. Not live Global-market coverage, background alerts, broker execution, and model training.
Decision boundary Historical and general informational research only. No personalized investment or trade advice, price target, trade execution, position-size output, or performance promise. Discovery supports review; it does not replace a person’s judgment or a regulated professional.
Turn a complex research, technical, or leadership career into a source-linked factual record that counsel can inspect. OnlineNIW connects each material claim to its exhibits, flags contradictions, and keeps reviewer responsibility explicit.
01Signal scanFind the claims the record can actually support.
02Evidence buildMap every material statement to a source.
03Boundary reviewSeparate factual, client, recommender, and counsel decisions.
Clear boundary OnlineNIW is operated by Explore AI Inc., not a law firm. It does not provide legal advice or representation, decide eligibility, or guarantee any immigration outcome. Public examples are synthetic.
03Prospective impactFactual record · counsel review requiredReview gate
Explore AI presents
Photon47
Travel through Gut, Neuro, Spore, Bone, Gene, and Liminal Loop, then cross the Heart finale. Meet an unruly molecular cast and see how the same heroes transform across story, arena, strategy, and action modes.
We separate measured outcomes from interpretation, describe limitations, and revise public claims when the evidence changes.
Build to learn
Products, prototypes, and interactive environments make abstract research questions concrete enough to test.
Efficiency matters
Capability is more useful when it can run with less computation, lower latency, and wider access.
Make complexity legible
Interfaces should help people understand what a system knows, what it is trying, and where uncertainty remains.
Preserve human agency
We want adaptive systems to extend human judgment and creativity—not quietly erase meaningful choice.
07 Company
Built across disciplines, focused on what comes next.
Our work is informed by experience in production cloud systems, deep-learning infrastructure, multimodal reasoning, autonomous agents, and AI for science.
Explore AI Inc. was founded in 2026 to connect long-horizon research with things people can use, question, and experience.
