Summary4 min read

Podcast Summary: Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs

Episode: GTC Bonus: Redefining AI Privacy with Eiman Ebrahimi of Protopia AI
Host: Logan Lawler
Guest: Eiman Ebrahimi, CEO & Founder of Protopia AI
Date: March 21, 2026

Episode Overview

This special GTC Bonus episode dives deep into privacy advancements in AI workflows, focusing on how Protopia AI is redefining data protection for modern AI deployments. Host Logan Lawler welcomes Eiman Ebrahimi, CEO and founder of Protopia AI, to discuss inference privacy, secure agentic systems, and the novel Stained Glass Transform technology. The conversation sheds light on real-world deployments across government, enterprise, and financial sectors, outlining what makes Protopia’s approach unique and impactful for safeguarding sensitive data in high-performance computing environments powered by Dell Pro Precision and NVIDIA RTX PRO GPUs.

Key Discussion Points & Insights

1. Introduction to Protopia AI and Inference Privacy

[00:19 - 00:52]

Eiman’s Background:
Eiman is the CEO and founder of Protopia AI, focusing on an “inference privacy layer for AI agents, AI factories,” enabling high utilization and secure multi-tenant compute on inference backends.
Main Offering:
Protopia AI’s technology acts as a privacy shield, allowing AI agents to function without exposing sensitive user data to external or hosted models.

2. Demystifying the Privacy Layer

[00:52 - 02:17]

Visual Explanation:
Eiman explains the “SafeGlaw” visual—a playful reference from their recent NVIDIA developer livestream, symbolizing privacy for agentic AI systems.
Challenge Addressed:
The industry often assumes the only way to keep data private is to localize models (e.g., large language models). Protopia challenges this by enabling agents to utilize hosted models without transmitting sensitive information in its original form.
Practical Example:
If an AI agent is scanning your emails to surface important ones, Protopia’s approach means you get the benefit without your raw emails ever leaving your device in readable form.

“That classification of what email you need to know about doesn’t need to expose your emails to the hosted model. And so that’s what our product enables.”
— Eiman Ebrahimi [01:30]

3. How Stained Glass Transform Works

[02:17 - 03:47]

Clarification on Technology:
Not traditional encryption:
- Encryption usually implies keys and reversible processes.
- Stained Glass Transform is a one-way function—no keys, no decryption.
Technical Process:
- Transforms plaintext tokens into “stochastic or randomized embeddings.”
- These embeddings are not reversible to their original text, yet still usable by the target model for inference.
- The process is machine-learned for compatibility with specific models.
“The thing you are sending are those random representations which are not reversible. There is no reverse function to go back to the original plaintext, but the target model will still be able to operate based on it.”
— Eiman Ebrahimi [03:20]

4. Industry Adoption and Use Cases

[03:47 - 04:57]

Federal and Government:
Immediate need for privacy-preserving AI solutions in federal and state applications.
State and Local Government (SLED) Spotlight:
Recently announced customer, management consulting firm MGT, is deploying an AI experience center to enable state & local workflows—without sharing raw customer data with managed services.
Enterprise and Application Providers:
- Protopia is used by AI application providers serving enterprises.
- E.g., Provost—a financial services analysis provider for high-net-worth families—that uses Protopia to deliver insights without aggregating customers’ private data.
Memorable Quote:

“They’ve stood up now an AI experience center that is going to enable workflows for state and local without needing those customers to share their plain raw information with the managed service.”
— Eiman Ebrahimi [04:20]

5. Learning More & Getting Involved

[04:57 - 05:40]

How to Connect:
- Main site: protopia.ai
- LinkedIn: Search ‘Protopia AI’
- NVIDIA Developer Livestream: Recently featured demo with broader technical and community engagement, plus a raffle for an NVIDIA DGX Spark.
- Partners and developers can submit requests for support on additional models or hosting systems.
“This particular NVIDIA livestream that we did on Wednesday is a really interesting place to learn more details and we have in there a way for you to also communicate what models you might want to see this sort of inference privacy layer for or what hosting systems you might want to see it for.”
— Eiman Ebrahimi [05:17]

Notable Quotes & Memorable Moments

On Protopia’s Impact:

“What we’re showing using our privacy preserving technology…is that you can have your OpenClaw agent talk to a hosted model without needing to actually send the sensitive information in plain text to that hosted model.”
— Eiman Ebrahimi [01:28]
On the Nature of Stained Glass Transform:

“The technology is not encryption… Here the stained glass transforms…they’re one way functions. There are no keys involved and you go from raw plaintext tokens to what we call stochastic or randomized embeddings.”
— Eiman Ebrahimi [03:00]

Important Timestamps

00:19: Introduction to Eiman Ebrahimi & Protopia AI
01:09: Privacy & security in agentic AI systems
02:54: Deep dive on Stained Glass Transform vs. traditional encryption
03:59: Real-world adoption: government, SLED, enterprise
05:09: How to connect & learn more

Summary Takeaway

Eiman Ebrahimi and Protopia AI are at the forefront of solving AI privacy challenges, enabling businesses and governments to leverage hosted, high-performance AI models—powered by Dell Pro Precision and NVIDIA RTX PRO GPUs—without exposing sensitive data. Their innovative Stained Glass Transform is reshaping how privacy can be embedded at the inference layer, moving beyond encryption and towards practical, non-reversible data transformations. The episode offers an accessible yet thorough dive into next-gen privacy solutions for AI workflows, showcasing disruptive potential across industries.

Loading summary

Transcript13 lines

[00:00]
A
Foreign. Welcome to reshaping workflows with Dell Pro Precision and Nvidia, where innovation meets real world impact in high performance computing.
[00:20]
B
That's Logan GTC lost the day 2026. Not even sure anymore. I'm here with Iman of Protopia AI. So Iman, give us a little background who you are, what you do at Protopia AI and just a brief overview of the company.
[00:35]
C
Thank you Logan. My name is Iman. I'm the CEO and founder at Protopia AI. We are an inference privacy layer for AI agents, AI factories and we enable high utilization multi tenant compute on inference backends.
[00:53]
B
Okay, so you've said a lot of words, let's break those words down. So let's talk a little bit. I see a dancing claw kind of in the background. Everyone loves crustaceans. Tell us a little bit about. Let's start with the privacy piece. What does Protopia AI do from a privacy and security standpoint for AI deployments?
[01:09]
C
Yeah, so the little visual you're seeing in the back, we call it SafeGlaw. We actually did a live stream on Nvidia's the developer livestream on Wednesday last week about this where we introduced the notion that in Open Claw Agentic systems a lot of people talk about the only way for keeping data private is to try to host the actual language models that the agent's going to try to talk to locally. But what we're showing using our privacy preserving technology that we call Stained Glass Transform, which is what jacket the claw is wearing there, is that you can have your OpenClaw agent talk to a hosted model without needing to actually send the sensitive information in plain text to that hosted model. So let's say your Agentix system wants to go through your emails and find things that it's going to notify you of. That classification of what email you need to know about doesn't need to expose your emails to the hosted model. And so that's what our product enables. Stained Glass Transform is something that runs on your local agent in order to connect to the hosted model. So that's an example of privacy for agentic AI using hosted models.
[02:17]
B
Okay, so we've got kind of a follow up, right? So say, you know, I don't know if I would let OpenClaw have access to my corporate emails. That's another, another thing. But let's say I did. Like we're kind of talking about, you know that point, like is there is an encryption, like what do you actually like the layer in between? Like what are you exposing to Ultimately give the agent and that, that that host. Well, I'm not hosted model but like let's say you're using like ChatGPT, what is it? What's the newest one. 5.2 or whatever you've opened that you have it connected to an API. Like what are you doing to protect that data? Is it limiting the data that's sent? I mean you don't have to share any like proprietary information. I'm just more curious.
[02:54]
C
No, that's a great question. So the technology is not encryption because encryption would imply that there are keys and there is a reverse function to go back from the encrypted data to the original data. Here the stained glass transforms that are created for target models, they're one way functions. There are no keys involved and you go from raw plaintext tokens to what we call stochastic or randomized embeddings. And those randomized embeddings are created in a way that the target model will still be able to operate accurately. And so the thing that you are sending are those random representations which are not reversible. There is no reverse function to go back to the original plaintext, but the target model will still be able to operate based on it. And those randomized transformations are learned, they are machine learning models themselves and they're learned for a given target model.
[03:47]
B
Okay, so give us for those that have listening. Never heard of Protopia. Like who are some of the. You don't have to name names, but like some of the industries or customers that you work with, where you've seen, you know, adoption of your product.
[03:59]
C
Yeah, for sure. So there's three or four main categories obviously as you can imagine in the federal space and in the government space, this is something that's high priority. So we do have customers in the federal space. We recently actually at GTC yesterday announced a new customer that is in the sled space. So it's a management consulting firm called MGT that provides customized solutions for state and local type customers and they've stood up now and an AI experience center that is going to enable workflows for state and local without needing those customers to share their plain raw information with the managed service and then also in the AI application provider space, especially those that work with enterprises. So we've had a couple of customer spotlights here. Provost is one company that does financial services kind of analysis for high net worth families, but again doesn't need to incorporate the information in order to bring that service to them.
[04:58]
B
All right, that's amazing. So you've heard it here first. Iman basically where can, if somebody's listening, you know, where can they find information about protopiai? What's the website? Where can they connect with you? What's the best way to get in contact with you?
[05:10]
C
Yeah, so Protopia AI is the website on LinkedIn. You can also find us under Protopia AI. This particular Nvidia livestream that we did on Wednesday is a really interesting place to learn more details and we have in there a way for you to also communicate what models you might want to see this sort of inference privacy layer for or what hosting systems you might want to see it for. And also there's a raffle for an Nvidia DGX Spark on there. So we're actively expanding our partnerships right now.
[05:40]
B
Amazing. Protopiai. Check it out. This is Logan live from gtc. And we're out.
[05:53]
A
Do what you want. Do what you want. This podcast was produced in partnership with Amaze Media Labs.