Co-Founder & CEO
Hamed Majidifard
PhD engineer and repeat founder with experience commercializing AI and computer vision for real-world engineering.
LinkedInBuilding AI that creates its next capability Toward artificial superintelligence for science and engineering
We are developing self-directed intelligence intended to form goals, acquire permitted resources, and create the specialist models, agent-native software, or complete systems its next objective requires.
Human-defined purposePurpose · permissions · limits
Self-directed intelligence
Designed to create its next capability
When the missing capability is structural, create a new capability rather than repeat the old behavior harder.
The long-term direction has five visible properties: set and revise goals inside a human-defined purpose; compose missing expertise; obtain permitted knowledge and resources; act, adapt, and recover across long horizons; and create new capabilities that can strengthen a later generation of the system.
The missing capability becomes the next thing the system creates
A scientific objective can demand expertise, a model, software, or a complete agent system that does not exist yet. Our direction is to make that absence actionable.
From applying fixed intelligence to creating new intelligence.
Scientific or engineering objectiveCapability not yet available
Complete specialist systemGenerate the framework and its specialists as one complete system rather than deploying a fixed roster of generic agents.
The long-term capability-creation program distinguishes six outputs: a specialist worker; a domain or subdomain model; a task-specific model; a complete AI model; agent-native software; and a complete specialist multi-agent system.
A model for the discipline. Another for the task
The field should determine the expertise—not the shape of the intelligence. Discovery, diagnosis, design, and invention should call for different intelligence, not different companies.
Domain experts should bring the objective without becoming AI-system operators.
Specialist portfolioCompose field and task specialists around the objective instead of forcing one enormous general model to imitate every form of expertise.
The long-term specialist-intelligence strategy separates a model for a field, a model for a subdiscipline, a model for a recurring technical capability, and a portfolio that composes those specialists around an objective.
Stop making agents operate software built for eyes and hands
An interface added to human software does not make it agent-native. Our direction is software whose native operator is an agent—and a system capable of creating that software, not merely consuming it.
The target is domain capability, not another programming helper.
Complete systemGenerate the framework, specialist intelligence, and machine-operable software as one coherent system for a field or objective.
The agent-native direction distinguishes software created for agent operation, reusable capability that packages how a scientific or engineering domain actually works, and complete systems whose framework and specialists are generated together.
The next move should not require the next prompt
Within a human-defined purpose, self-direction means generating the next instruction, obtaining what is missing, continuing when no person is present, repairing failure, and changing the available capabilities when repetition is not enough.
Human-defined purposePurpose · permissions · resource limits · consequential authority
01 · Long-term direction
Derive goals and self-prompt
Derive and revise the next goal inside a human-defined purpose—not wait for a human to author every prompt.
- Set next goal
- Generate instruction
- Revise direction
The long-term self-direction program has six responsibilities: set and revise goals and generate the next prompt; compose specialist intelligence; acquire permitted knowledge and resources; act autonomously, adapt, and complete missing work; test itself, diagnose failure, heal, and retry; and create new capabilities that can strengthen later generations.
Autonomy cannot scale on frontier-model economics
Long-horizon systems cannot rely on a frontier-scale generalist for every action. Intelligence must become smaller, faster, and purpose-built, while scarce frontier compute is reserved for decisions that truly require it.
Move human expertise from relaying work between machines to deciding what is worth pursuing.
Purpose-built intelligenceUse small domain and task specialists where they are sufficient, and reserve frontier-scale intelligence for the decisions that need it.
Research objective Lower inference burden and latency so sustained autonomous work becomes economically possible.
The economics thesis contrasts using one frontier-scale general model for every action with composing smaller domain and task specialists and reserving expensive general intelligence for decisions that require it.
Each generation should build a stronger next generation
Recursive improvement is not another model learning from conversation. The intended system redesigns the workers, models, software, and specialist systems available to its successor.
Eventually: design, build, train, test, and evaluate a complete new AI model end to end.
Generation n · Builds its successorThe system creates and incorporates stronger models, software, and specialist systems for the generation that follows.
Parallel research direction Models that adapt while in service instead of remaining frozen between occasional retraining events.
The recursive-succession direction progresses from a system using available intelligence, to creating a missing specialist, to creating models and agent-native software, and ultimately to building a more capable successor system.
Artificial superintelligence for science and engineering is a system property—not a bigger-model label
The destination is an intelligence able to form technical goals; acquire the knowledge and permitted resources its next objective requires; create the needed models, agent-native software, and specialist systems; act across long horizons; and improve the system that created them.
Forms + revises goals
Acquires permitted resources
Creates specialist intelligence
Builds agent-native systems
Improves its successor
Long-term horizonASI for science
and engineering
and engineering
The horizon is defined by behavior: self-directed, specialist, agent-native, economically scalable, and capable of creating stronger intelligence.
Human authority remains explicit: purpose, permissions, resource limits, and consequential decisions.
Technical founders across AI, robotics, and engineering
Singulus AI is led by researchers and operators across applied AI, computer vision, robotics, and engineering.
Co-Founder & CTO
Ali Shafiekhani
PhD computer scientist with published and patented work in robotics perception and computer vision.
LinkedInHelp shape self-directed intelligence for science and engineering
We want to speak with researchers, engineers, model builders, and institutions thinking beyond single-agent automation.
Start a conversation