This roadmap is how micro1 becomes the data infrastructure for AGI.
At micro1, we are building the systems that determine who produces intelligence, how it is evaluated, and how it compounds into real model capabilities.
As this infrastructure advances over time, it also creates a powerful byproduct: the largest expert network and the largest new category of work created by the AI economy.
This documentation reveals the details about each individual platforms, products, and features that contribute to this mission.
.webp)
.svg)
.webp)
.svg)
.webp)
What we’ve built
micro1 is currently built on two tightly integrated platforms
Data Engine
The underlying infrastructure that produces the high-quality expert data powering the intelligence platform. While not directly revenue-generating, it is foundational to everything we ship.
Intelligence Platform
World-class data products used by frontier AI labs, enterprises, and robotics companies to train and deploy cutting-edge models.
Here is how each platform is broken down
.svg){kind=small}
The micro1 data engine
-min.png)
Zara
At the center of human data is humans, and Zara is the core sourcing and vetting system that powers the micro1 data engine. It recruits domain experts with extremely high accuracy and speed, forming the human foundation that generates net-new expert data for AI labs and enterprises. The Zara product has 4 main components, as described below.
Referral program
The primary sourcing channel for expert talent. Existing experts are paid anywhere from $100 to $2,000 per successful referral, creating a self-reinforcing supply loop. As more top-tier experts participate in paid work, the referral pool expands, improving both scale and candidate quality over time.
Skill-based AI interviewer
Once sourced, 2,000+candidates a day go through an AI-led interview that evaluates domain-specific skills through live conversation. This process also includes a human data exercise that closely simulates real on-the-job work. The system is powered by multiple proprietary components, including the conversational m1 model, custom proctoring model, and our on-the-fly exercise generation model
m1 blueprint
An internal operational tool used to select candidates, send offers, and onboard talent at scale.
Talent dashboard
After passing the AI interview, experts complete onboarding through the talent dashboard and become micro1-certified. The dashboard allows experts to onboard for payouts, see total earnings, access our upskilling courses, apply to new jobs, and more.
-min.png)
Data platform
The environment where domain experts create, review, and ultimately submit complex datasets. This is the second major pillar of the data engine and consists of two components:
Human data platform
The primary sourcing chThe core data production infrastructure for all pipelines and customers at micro1. It supports a wide range of expert data creation workflows including rubric-based expert data, VLM annotation, robotics video creation, and other multimodal expert tasks.annel for expert talent. Existing experts are paid anywhere from $100 to $2,000 per successful referral, creating a self-reinforcing supply loop. As more top-tier experts participate in paid work, the referral pool expands, improving both scale and candidate quality over time.
Rhea
AI-based quality control system trained to detect structural errors in expert-generated data. Rhea learns from project instructions and serves as the first layer of multi-model QC across environments.
Talent Performance Management Dashboards
Our performance management dash-boarding system that continuously quantifies expert quality, velocity, and reliability. It creates clear, consistent performance signals that inform both bonuses and replacements, while feeding back into vetting and QA systems to improve data quality over time.
The micro1 Intelligence platform
-min.png)
This is the human intelligence layer that transforms expert data into frontier model improvements, enterprise agents, and real-world robotics capability. The intelligence platform consists of four core products
Realm
Expert-level human data for training and improving frontier foundational models. Realm includes RL environments and frontier evaluations designed to drive domain-specific capability improvements.
Cortex
Contextual human data used to evaluate, fine-tune, and improve enterprise AI agents. Cortex generated real-world workflows and niche expert enterprise data.
Robotics
Real-world robotics data for physical intelligence. This vertical includes three pipelines: human demonstrations, teleoperation, and VLA data. Together, these pipelines are designed to create the world’s largest and highest quality real-world robotics dataset.
Data engine API
A simple interface that allows customers to request data, receive structured outputs (e.g. JSON), and approve results across AI lab, enterprise, and robotics use cases.
What's next:
The micro1 roadmap
.webp)
.webp)
Robotics tasks app
March 2026
A real-world robotics data collection app focused on human demonstration. It will allow anyone, anywhere, to generate physical intelligence data using their phone by recording tasks they already perform at home. Contributors are paid to generate data while helping train friendly robots that will assist with those same tasks in the future.
Short-term: a major boost to the human demonstration pipeline via crowdsourced data.
Long-term: the single largest robotics dataset ever created and one of the highest-earning gig platforms globally.
Evals tab
March 2026
An evaluation and production monitoring platform for enterprise agents. Enterprises can upload agent functions and data consumption limits to receive automated evaluation charts across all functions within days. Over time, the evaluation framework improves its accuracy and provides targeted suggestions for agent improvements and fine-tuning data recommendations. The Evals tab, within the Intelligence platform, will also include a leaderboard of top foundation models across micro1 RL environments, and a robotics leaderboard for physical intelligence performance.
Expert-level human data for training and improving frontier foundational models. Realm includes RL environments and frontier evaluations designed to drive domain-specific capability improvements.
Long-term: the single largest robotics dataset ever created and one of the highest-earning gig platforms globally.
Reward Model as a Service (RMaaS)
May 2026
As micro1 accumulates large-scale, non-exclusive datasets across Realm, Cortex, and Robotics, it becomes possible to train domain-specific reward models grounded in real expert behavior rather than synthetic benchmarks. RMaaS provides access to these reward models via API, allowing customers to directly improve policy models or agents using micro1-built environments. Examples:
In Realm, a W-2 U.S. taxation RL environment populated with expert tax data across all edge cases would produce the best reward model for taxation. Any model seeking strong taxation performance would need to subscribe to micro1’s RMaaS API.
The same applies to robotics: a real-world vacuuming environment could yield the best physical intelligence reward model for that task.
Over time, RMaaS becomes the final infrastructure layer for foundation models in specific domains.
The large-scale training and fine-tuning of AI models is giving rise to one of the largest markets humanity has ever created, driven by the need for continuous, high-quality human oversight to train, evaluate, and align intelligent models. A combination of micro1's current product suite and near-term roadmap collectively form the infrastructure required to operate within this market at scale by sourcing expert intelligence, capturing it inside real workflows, measuring performance over time, and converting that signal into RL environments, evaluations, and reward models that directly compound model capability. As this market takes shape, the platforms that establish these systems early will define how intelligence is trained, priced, and scaled for decades to come.
