CTGT wins Best Presentation Style award at VB Transform 2025

Based in San Francisco CTGTA startup focused on creating more reliable AI thanks to the personalization of the model in terms of functionalities, won the prize for the best style of presentation VB Transform 2025 in San Francisco. Founded by Cyril Gorlla, 23, the company has shown how its technology helps companies overcome AI confidence barriers by directly modifying model functionality instead of using traditional fine or intensity engineering methods.

During her presentation, Gorlla underlined the “Doom AI loop” confronted with many companies: 54% of companies cite AI as their highest technological risk according to Deloitte, while McKinsey reports that 44% of organizations have experienced negative consequences of the implementation of AI.

“Much of this conference was on the AI ​​Doom loop,” said Gorlla during his presentation. “Unfortunately, many of them [AI investments] Do not take place. I have just been canceled Hundreds of AI pilots because they have not really delivered a return on investment because of no fundamental confidence in these systems. »»

Break the AI ​​calculation wall

CTGT’s approach represents a significant difference compared to conventional AI personalization techniques. The company was based on Research Gorlla carried out while holding a chair endowed at the University of California San Diego.

In 2023, Gorlla Published an article During the International Conference on Representations of Learning (ICLR) describing an assessment and training method of AI models which exceeded up to 500 times faster than existing approaches while carrying out “three nine” (99.9%) of precision.

Rather than relying on gross scaling or traditional methods of in -depth learning, CTGT has developed what he calls an “entirely new AI pile” which fundamentally reinvents the way in which neural networks learn. The innovation of the company focuses on understanding and intervention in terms of the functionality of AI models.

The company’s approach differs fundamentally from standard interpretation solutions based on secondary AI systems for surveillance. Instead, CTGT offers mathematically verifiable interpretation capacities which eliminate the need for additional models, which considerably reduces calculation requirements in the process.

The technology works by identifying specific latent variables (neurons or directions in the space of characteristics) which cause behavior such as censorship or hallucinations, then dynamically modifying these variables at the time of inference without modifying the weights of the model. This approach allows companies to personalize the behavior of the model on the fly without removing offline systems for recycling.

Real world applications

During her transformation presentation, Gorlla demonstrated two business requests already deployed in a financial institution in fortune 20:

An e-mail compliance work flow which forms models to understand the acceptable content specific to the company, allowing analysts to check their emails compared to compliance standards in real time. The system highlights potentially problematic content and provides specific explanations.

A brand alignment tool that helps marketing specialists to develop a coherent copy with brand values. The system can suggest personalized advice on the reasons why certain sentences work well for a specific brand and how to improve the content that does not align.

“If a company has 900 use cases, it no longer has to refine 900 models,” said Gorlla. “We are agnostics, so that they can simply connect us.”

A real example of CTGT technology in action was his work with Deepseek modelswhere he successfully identified and modified the features responsible for censorship behavior. By insulating and adjusting these specific activation models, CTGT was able to reach a 100% response rate on sensitive requests without degrading the performance of the model on neutral tasks such as reasoning, mathematics and coding.

Images: Presentation CTGT to VB Transform 2025

Demonstrated king

CTGT technology seems to provide measurable results. During the question and answer session, Gorlla noted that during the first week of deployment with “one of the main insurers supplied by AI, we saved $ 5 million of their responsibility”.

Another first customer, Ebrada Financial, used CTGT to improve the factual accuracy of customer service chatbots. “Previously, hallucinations and other errors in the Chatbot responses led a high volume of live support agents while customers were trying to clarify the responses,” said Ley Ebrada, founder and tax strategist. “CTGT has helped to improve the accuracy of the chatbot enormously, eliminating most of these agent requests.”

In another case study, CTGT worked with a Fortune 10 Sans Name company to improve AI’s capacities on devices in constrained calculation environments. The company has also helped a leading computer vision company to obtain faster 10x model performance while maintaining comparable precision.

The company claims that its technology can reduce hallucinations by 80 to 90% and allow AI deployments with a reliability of 99.9%, a critical factor for companies in regulated industries such as health care and finance.

From Hyderabad to Silicon Valley

Gorlla’s journey is himself remarkable. Born in Hyderabad, India, he Mastered coding At 11 and dismantled laptops in high school to express more performance for the training of AI models. He came to the United States to study at the University of California in San Diego, where he received the Equipped Chair Stock Exchange.

His research focused on understanding the fundamental mechanisms of the way in which neural networks learn, which led to his article ICLR and finally CTGT. At the end of 2024, Gorlla and the co-founder Trevor Tuttle, an expert in hyperscalable ML systems, were selected for the Fall 2024 lot of Combinator.

The startup has attracted notable investors beyond its institutional donors, including Mark Cuban and other eminent technology leaders attracted by its vision of making AI more efficient and trustworthy.

Financing and future

Founded in mid-2024 by Gorlla and Tuttle, CTGT lifted $ 7.2 million In February 2025, in a rounded seed tower led by Gradient, the IA fund at the early stage of Google. Other investors include General Catalyst, Y Combinator, Liquid 2, Deepwater and notable angels such as François Chollet (Creator of Keras), Michael Seibel (Y Combinator, co-founder of Twitch) and Paul Graham (Y Combinator).

“The launch of the CTGT is appropriate, because the industry is struggling to develop AI within the current limits of computer limits,” said Darian Shirazi, director partner at Gradient. “CTGT removes these limits, allowing companies to quickly set their AI deployments to scale and execute advanced AI models on devices such as smartphones. This technology is essential to the success of AI deployments with high issues in large companies. ”

The size of the AI ​​model exceeding the law and the progress of Moore in IA training fleas, CTGT aims to focus on a more fundamental understanding of AI which can face both ineffectiveness and increasingly complex model decisions. The company plans to use its seed financing to extend its engineering team and refine its platform.

Each finalist has presented himself to an audience of 600 industry decision-makers and received comments from a panel of Judges de Capital-Résque de Salesforce Ventures, Menlo Ventures and Amex Ventures.

Read the other winners Catio and solo.io. The other finalists were Fist,, Superduper.io,, Sutra And Qdrant.

Publisher’s note: as thanks to our readers, we have opened the recording of highlights for VB Transform 2026 – only $ 200. This is where the ambition of the AI ​​meets operational reality, and you will want to be in the room. Book your place now.