Santosh Kumar Radha is the co-founder and chief technology officer of AgentField, an open-source platform for coordinating autonomous AI agents. A theoretical physicist turned infrastructure entrepreneur, he previously helped build Covalent, the distributed-computing platform acquired with its parent company, Agnostiq, by DataRobot in 2025.
Radha earned a doctorate in theoretical physics from Case Western Reserve University, studying massive gravity and condensed-matter physics before pursuing practical quantum algorithms. His published research includes quantum financial-option pricing and constraint learning for quantum optimization.
At Agnostiq, he led quantum-algorithm work and later assumed responsibility for product and engineering. Covalent brought scientific, quantum, and machine-learning workloads into a common Python-based orchestration system spanning different computing environments. His 2024 AI Engineer demonstration showed how notebook functions could fine-tune models on remote GPUs, evaluate candidates, shift lightweight comparisons to CPUs, and deploy an autoscaling inference endpoint without developers managing Docker or Kubernetes.
DataRobot acquired Agnostiq in February 2025. Radha subsequently co-founded AgentField with Oktay Goktas, extending his orchestration work from heterogeneous compute to autonomous-agent systems.
- Agent control planes: AgentField treats autonomous agents as production services requiring identity, coordination policies, verifiable execution histories, and interoperability.
- Self-assembling agent meshes: Specialist agents register their capabilities, allowing other agents to discover and invoke tools dynamically without coordinated redeployments.
- Runtime-shaped workflows: Execution graphs form around the task: a reasoning component selects necessary specialists, runs suitable work concurrently, and preserves an execution record.
- Failure-aware coding agents: Autonomous software-development systems separate inexpensive routing from resource-intensive coding environments while using checkpoints, differentiated retries, testing, and explicit reporting of unresolved defects.
Radha also collaborates with Goktas on UWM-JEPA, which explores representing uncertainty in AI world models through density-matrix latent states and learned unitary evolution.
