CTGT wins Best Presentation Style award at VB Transform 2025

San Francisco based CtgtThe feature level has won the best presentation style award, a start of AI to be more reliable through model customization VB Transform 2025 In San Francisco. The company established by 23-year-old Cyril Gorlla demonstrated that technology has helped the AI ​​True obstacles to repel the AI ​​trust obstacles instead of using traditional subtle regulation or emergency regulation methods.

During the presentation, Gorlla stressed the “AI Doom Loopu”, which faces many enterprises: 54% of enterprises are the highest technological risks of the AI, 44% of organizations express negative results of the AI ​​application.

“A large part of this conference is about AI doom loop,” Gorlla, described during the presentation. “Unfortunately, most of these [AI investments] Don’t catch. J & J was only canceled Hundreds of AI pilots because they did not deliver ROI due to the main trust in these systems. “

Break the AI ​​billing wall

The CTGT approach represents an important departure of ordinary AI customization techniques. The company was also established in San Diego University of the University of California, the research was established in Gorl.

Gorlla in 2023 A paper was published AI models 500 times faster than the existing approaches in the International Conference on Training Representations (ICLR), 500 times faster than the existing approaches are described in the assessment and training method.

CTGT has developed a “entirely new AI stack” how the CTGT studied how the CTGT is learning how the Neural networks learned how the Neural Neural networks learned. The company’s innovation is aimed at understanding and interference in the level of the AI ​​models.

The company’s approach is based on standard interpretation solutions that trust the secondary AI systems for monitoring. Instead, the CTGT offers a mathematical transcriptive ability to verify the need for significantly reducing the requirements of the billing requirements in the process, which eliminates the needs of additional models.

Technology, censorship or hallucinations, identifying the specific secret variables (feature fields in the field), such as censorship or hallucinations (feature neurons or directions), then change these variables in dynamically. This approach allows companies to develop model behavior without drawing systems for retraining.

Real-world apps

During the presentation of the transformat presentation, Gorlla has already demonstrated two enterprises placed in Fortune 20 financial institutions:

Models to understand the specific acceptable content of the models, an email compatibility workflow that allows an analytics to inspect their email against real-time compatibility standards. The system highlights potentially problematic content and presents concrete explanations.

Marketers are a brand adaptation tool that helps develop a copy of the brand values. The system may offer individual tips on how to improve why specific expressions are working well for a specified brand and how to improve the non-aligned content.

“If a company is 900 uses, they do not have to make 900 models,” Gorl said. “We are model-agnostic, so they can join us.”

Was an example of the real world of CTGT’s technology DeepSeek modelssuccessfully identified and identified the features responsible for censorship behavior successfully. It was able to reach 100% response in sensitive inquiries without humiliating the performance of the model performance on neutral tasks such as CTGT, justification, mathematics and coding, as to isolate and adjusting special activation samples.

Images: CTGT presentation in VB Transform 2025

Demonstrated

CTGT technology seems to be measurable results. Gorlla noted during the Q & A meeting that “in the first week of deploying the leading AI-power with one of the insurers, we maintained $ 5 million.”

Another early client, Ebrade Financial, Customer Service used CTGT to increase the actual accuracy of ChatBots. “Previously, ChatBot respondings required a high volume for live support agents, the founder and tax strategy for living support agents, because they want to clarify the answers,” said Ley Ebra. “CTGT helped to improve the accuracy of chatbot by eliminating the majority of the agent of the agent.”

In another study investigation, CTGT has worked with a Fortune 10, which has no name to increase AI capabilities on the device in the report-contour environment. The company also helped to achieve 10x faster model performance of the leading computer vision while maintaining comparable accuracy.

The company claims that its technology can reduce 80-90%, and a critical factor of enterprises in regulated areas such as 99.9% reliability, EU placements, health and finance.

From Hyderabad to Silicon Valley

Gorlla’s travel itself is noteworthy. Hyderabad, India, O Masturous coding 11 years old and ripped luxurious laptops to squeeze more performance for AI models. At the University of California, studying at San Diego University, studying at San Diego University, received a scholarship.

His research there was aimed at understanding how the neural networks learned how the neural networks, and eventually caused the CTGT. In late 2024, Gorlla and co-founder Trevor Telttle, Y Combinator’s Autumn 2024 party were elected in the end of 2024.

Starting, Mark Cuba and other prominent technology leaders, including AI more effective and reliable, and other famous technology leaders also attracted attractive investors.

Financing and future

In mid-2024 Gorlla and Tuttle, established by CTGT Raised $ 7.2 million In February 2025, the early seed tour of the gradient, Google’s early stage Ai Foundation. Other investors include highest catalytics, Y combinations, liquid 2, deep water and franchois chollet (keras creator), Michael Seibel (Y Compinator, Combinator) and Paul Graham (Y Combinator) and Paul Graham (Y Combinator).

“CTGT is struggling with the scale of the AI ​​in the current limits of the industrial and calculation limits,” he said. “CTGT removes these limits and allows companies to quickly scalp the AI ​​placements and launch advanced AI models on devices such as smartphones. This technology is important for the success of high stakes in large enterprises.”

With AI model size, Moore’s law and AI training hips aims to focus on the basic concept that can cope with advances, CTGT, both inefficiency and increasingly complex model decisions. The company plans to use seed funding to expand its engineering group and clean the platform.

Each finalist presented 600 industrial decision makers and received an opinion from the panel of ventures from Salesforce Venturas, Menlo Enterprises and AMEX enterprises.

Read about other winners Castio and solo.io. Were other finalists Punch, Superduper.io, Sutra and Grant.

Editor’s note: Asked our readers, we opened an early bird registration for VB Transform 2026 – only $ 200. This is ai fame, and you will want to be in the room in the room. Protect your place now.