In this episode, we sit down with Brian Geisel, Founder of Symage from Geisel Software, to explore how synthetic data, robotics, machine learning, and large language models (LLMs) are reshaping AI infrastructure.
About This Episode
Synthetic data is rapidly becoming one of the most important building blocks in AI, from training autonomous vehicles to powering robotics on Mars. John Kosturos sits down with Brian Geisel, Founder of Symage from Geisel Software, to explore how synthetic data, robotics, machine learning, and large language models are reshaping AI infrastructure and why generating better, targeted data matters far more than generating more of it.
Brian Geisel
LinkedInFounder, Symage at Geisel Software
Brian is the founder of Symage, a synthetic data platform built under Geisel Software. With deep roots in machine learning, robotics, and AI systems engineering, Brian has worked on projects spanning NASA Mars rover simulation environments to warehouse automation and physical AI. He brings a practitioner’s lens to the often misunderstood world of synthetic data generation and its role in training the next generation of AI models.
How NASA Trained Mars Rovers Using Synthetic Environments
Why real-world data collection is impossible for some of the most demanding AI applications, and how synthetic simulation environments solved that problem for NASA’s Mars missions.
Why Humanoid Robots May Not Be the Future
Brian’s contrarian take on the humanoid robotics hype cycle, and what purpose-built physical AI designs may be better suited for the real problems enterprises actually need solved.
Robotics, Warehouse Automation, and Micro-Fulfillment
How robotics trained on synthetic data is transforming warehouse operations and micro-fulfillment centers, and what the infrastructure demands of physical AI look like at scale.
Why Foundation Models Must Rely on Synthetic Data
The internet has been scraped. Real-world data is running out. Brian explains why synthetic generation is no longer optional for foundation model training and what that means for the next generation of LLMs.
The Hidden Storage Challenges Behind AI Training
Synthetic data pipelines generate massive volumes of structured and unstructured data. Brian unpacks the storage architecture decisions that most teams get wrong and the cost implications of getting it right.
How Synthetic Data Solves PII and Regulated Industry Problems
Healthcare, finance, and other regulated sectors cannot freely use real patient or customer data for training. Synthetic data offers a privacy-preserving path that maintains statistical fidelity without compliance risk.
Balancing Real-World Data with AI-Generated Data
Synthetic data is not a wholesale replacement for real data. Brian explains the right ratio, when each type earns its place in a training pipeline, and how to avoid the model collapse trap.
GPU vs. Object Storage for AI Workloads
Why the conversation about AI compute cannot be separated from storage strategy, how high-throughput data movement becomes the bottleneck at scale, and what multimodal synthetic pipelines demand from infrastructure.
AI and ML Engineers Building Training Pipelines
If you’re designing or scaling a model training pipeline and have hit the ceiling on real-world data availability, Brian walks through how synthetic generation can fill the gaps without sacrificing model quality.
Robotics Engineers and Physical AI Teams
From warehouse automation to edge deployment, Brian’s firsthand experience with simulation-trained robotics systems offers practical insight into how physical AI and synthetic environments are converging.
Data Infrastructure and Storage Architects
Synthetic data generation creates unique storage and throughput challenges that most teams underestimate at the start. This episode gives you a realistic picture of what AI-scale data movement requires.
Healthcare, Finance, and Regulated Industry Leaders
Operating under HIPAA, GDPR, or other compliance frameworks? Brian explains how synthetic data is unlocking AI adoption in sectors where using real customer or patient data for training is simply not an option.
Investors and Founders in AI, Robotics, and Data Infrastructure
Synthetic data is still early and largely misunderstood as a category. This conversation maps out where the real value is being created, from foundation model training to edge robotics to regulated enterprise AI.
About Symage
Symage is a synthetic data platform developed by Geisel Software, designed to generate high-quality, targeted training data for machine learning models, robotics systems, and AI applications. By combining simulation environments, generative techniques, and domain expertise, Symage helps teams train better models faster while solving the data scarcity, PII compliance, and labeling cost challenges that slow down real-world AI deployments.
Learn more at geisel.software ↗
