Become a member of the event that is trusted by business leaders for almost two decades. VB Transform brings together the people who build the real Enterprise AI strategy. Leather
San Francisco established CTGTA startup aimed at making AI reliable by means of model adjustment at play level, won the best presentation style prize at VB Transform 2025 In San Francisco. Founded by the 23-year-old Cyril Gorlla, the company showed how its technology companies helps to overcome AI-Trust barriers by changing model functions directly instead of using traditional refinement or rapid engineering methods.
During his presentation, Gorlla emphasized the “AI Doom Loop” with which many companies are confronted: 54% of the companies mentions AI as their highest technical risk, according to Deloitte, while McKinsey reports that 44% of the organizations have experienced negative consequences of AI implementation.
“A large part of this conference was about the AI Doom Loop,” Gorlla explained during his presentation. “Unfortunately, many of these [AI investments] Pan not out. J&J just canceled Hundreds of AI pilots because they did not really deliver ROI because of no fundamental confidence in these systems. “
The AI Compute Wall Break
The approach of CTGT is a considerable deviation from conventional AI adjustment techniques. The company was founded on research Gorlla, while held a beneficiary chairman at the University of California San Diego.
In 2023, Gorlla a paper published At the International Conference on Learning Representations (ICLR) that describes a method for evaluating and training AI models that was up to 500 times faster than existing approaches, while “three nines” (99.9%) of accuracy was reached.
Instead of trusting brutal-force scaling or traditional deep learning methods, CTGT has developed what it calls a “completely new AI-stack” that fundamentally represents how neural networks learn. The innovation of the company focuses on understanding and intervention on the job level of AI models.
The approach of the company is fundamentally different from standard interpretability solutions that depend on secondary AI systems for monitoring. Instead, CTGT offers mathematically verifiable interpretability options that eliminate the need for additional models, which considerably reduces the calculation requirements in the process.
The technology works by identifying specific latent variables (neurons or directions in the characteristic room) that control behavior such as censorship or hallucinations and then dynamically modify these variables in inference time without changing the weights of the model. This approach enables companies to immediately adjust model behavior without taking systems offline for retraining.
Real-World Applications
During his transformation presentation, Gorlla demonstrated two operating applications that have already been deployed at a Fortune 20 Financial Instution:
A workflow for e-mail compliance that trains models to understand company-specific acceptable content, so that analysts can check their e-mails in real time for compliance standards. The system may emphasize problematic content and offers specific explanations.
A brand alignment tool that helps marketers to develop copies that are consistent with brand values. The system can suggest personalized advice on why certain sentences work well for a specific brand and how the content can be improved that does not line up.
“If a company has 900 use cases, they no longer have to refine 900 models,” Gorlla explained. “We are Model-Magnostic, so they can just join us.”
A Real-World example of CTGT’s technology in action was the work with Deepseek modelsWhere the functions responsible for the behavior of censorship are successfully identified and changed and changed. By insulating and adjusting these specific activation patterns, CTGT was able to reach a 100% response rate on sensitive questions without breaking down the performance of the model on neutral tasks such as reasoning, mathematics and coding.
Images: CTGT presentation at VB Transform 2025
Demonstrated ROI
CTGT technology seems to deliver measurable results. During the Q&A session, Gorlla noted that in the first week of deployment with “one of the leading AI-driven insurers we saved $ 5 million in liability of them.”
Another early customer, Ebrada Financial, has used CTGT to improve the actual accuracy of Chatbots for customer service. “Earlier, hallucinations and other errors in chatbot -answers made a large number of requests for live support agents, while customers tried to clarify answers,” said Ley Ebrada, founder and tax strategist. “CTGT has greatly improved the accuracy of the chatbot, so that most of those agent requests are eliminated.”
In another case Study, CTGT collaborated with an unnamed Fortune 10 company to improve AI options on the devices in computational limited environments. The company also helped achieve a leading computer vision company 10x faster model performance while maintaining similar accuracy.
The company claims that its technology can reduce hallucinations by 80-90% and enable AI implementations with 99.9% reliability, a critical factor for companies in regulated industries such as health care and finance.
From Hyderabad to Silicon Valley
Gorlla’s journey is remarkable. Born in Hyderabad, India, he Mastered Coding At the age of 11 and high school laptops disassembled to squeeze more performance for training AI models. He came to the United States to study at the University of California, San Diego, where he received the Fellowship of the Endowed Chair.
His research there was aimed at understanding the fundamental mechanisms of how neural networks learn, which led to his ICLR paper and ultimately CTGT. At the end of 2024, Gorlla and co-founder Trevor Tuttle, an expert in Hyperscalable ML systems, were selected for the autumn 2024 of Y Combinator.
The startup has attracted remarkable investors than its institutional financiers, including Mark Cuban and other prominent technology leaders who are attracted to his vision to make AI more efficient and reliable.
Financing and future
Mid-2024 Founded by Gorlla and Tuttle, CTGT $ 7.2 million picked up In February 2025 in an overvalued seed round led by Gradient, Google’s early stage AI Fund. Other investors are a general catalyst, Y-combinator, liquid 2, deep water and remarkable angels such as François Chollet (maker of Keras), Michael Seibel (Y Combator, co-founder of Twitch) and Paul Graham (Y Combator).
“The launch of CTGT is on time because the industry is struggling with scaling AI within the current boundaries of computer limits,” said Darian Shirazi, managing partner at Gradient. “CTGT removes those limits, so that companies can quickly scale up their AI implementations and perform advanced AI models on devices such as smartphones. This technology is crucial for the success of AI implementations with high commitment to large companies.”
With the AI model size that exceeds Moore’s law and the progress in AI training ships, CTGT wants to concentrate on a more fundamental understanding of AI that can deal with both inefficiency and increasingly complex model decisions. The company plans to use its seed financing to expand its engineering team and refine its platform.
Each finalist presented to an audience of 600 decision makers in the industry and received feedback from a panel of venture capital judges from Salesforce Ventures, Menlo Ventures and Amex Ventures.
Read about the other winners Catio and Solo.io. The other finalists were Fist” Superuper.io” Sutra And QDRANT.
Note of the editors: As a thank you for our readers, we have Early Bird Registration for VB Transform 2026 open-Slechts $ 200. This is where AI Ambition meets the operational reality, and you want to be in the room. Reserve your place now.
