CTGT wins Best Presentation Style award at VB Transform 2025

San Francisco-based CTGT, a startup centered on making AI extra reliable via feature-level mannequin customization, gained the Best Presentation Style award at VB Transform 2025 in San Francisco. Founded by 23-year-old Cyril Gorlla, the corporate showcased how its know-how helps enterprises overcome AI belief obstacles by instantly modifying mannequin options as a substitute of utilizing conventional fine-tuning or immediate engineering strategies.

During his presentation, Gorlla highlighted the “AI Doom Loop” confronted by many enterprises: 54% of companies cite AI as their highest tech danger in accordance with Deloitte, whereas McKinsey experiences 44% of organizations have skilled unfavorable penalties from AI implementation.

“A large part of this conference has been about the AI doom loop” Gorlla defined throughout his presentation. “Unfortunately, a lot of these [AI investments] don’t pan out. J&J just canceled hundreds of AI pilots because they didn’t really deliver ROI due to no fundamental trust in these systems.”

Breaking the AI compute wall

CTGT’s method represents a big departure from standard AI customization methods. The firm was based on analysis Gorlla carried out whereas holding an endowed chair on the University of California San Diego.

In 2023, Gorlla revealed a paper on the International Conference on Learning Representations (ICLR) describing a way for evaluating and coaching AI fashions that was as much as 500 instances sooner than present approaches whereas attaining “three nines” (99.9%) of accuracy.

Rather than counting on brute-force scaling or conventional deep studying strategies, CTGT has developed what it calls an “entirely new AI stack” that essentially reimagines how neural networks be taught. The firm’s innovation focuses on understanding and intervening on the function degree of AI fashions.

The firm’s method differs essentially from customary interpretability options that depend on secondary AI programs for monitoring. Instead, CTGT affords mathematically verifiable interpretability capabilities that remove the necessity for supplemental fashions, considerably decreasing computational necessities within the course of.

The know-how works by figuring out particular latent variables (neurons or instructions within the function house) that drive behaviors like censorship or hallucinations, then dynamically modifying these variables at inference time with out altering the mannequin’s weights. This method permits firms to customise mannequin conduct on the fly with out taking programs offline for retraining.

Real-world functions

During his Transform presentation, Gorlla demonstrated two enterprise functions already deployed at a Fortune 20 monetary establishment:

An e-mail compliance workflow that trains fashions to grasp company-specific acceptable content material, permitting analysts to verify their emails in opposition to compliance requirements in real-time. The system highlights probably problematic content material and supplies particular explanations.

A model alignment instrument that helps entrepreneurs develop copy in line with model values. The system can recommend customized recommendation on why sure phrases work effectively for a selected model and the way to enhance content material that doesn’t align.

“If a company has 900 use cases, they no longer have to fine-tune 900 models,” Gorlla defined. “We’re model-agnostic, so they can just plug us in.”

An actual-world instance of CTGT’s know-how in motion was its work with DeepSeek fashions, the place it efficiently recognized and modified the options accountable for censorship behaviors. By isolating and adjusting these particular activation patterns, CTGT was in a position to obtain a 100% response price on delicate queries with out degrading the mannequin’s efficiency on impartial duties like reasoning, arithmetic and coding.

Images: CTGT presentation at VB Transform 2025

Demonstrated ROI

CTGT’s know-how seems to be delivering measurable outcomes. During the Q&A session, Gorlla famous that within the first week of deployment with “one of the leading AI-powered insurers, we saved $5 million of liability from them.”

Another early buyer, Ebrada Financial, has used CTGT to enhance the factual accuracy of customer support chatbots. “Previously, hallucinations and other errors in chatbot responses drove a high volume of requests for live support agents as customers sought to clarify responses,” stated Ley Ebrada, Founder and Tax Strategist. “CTGT has helped improve chatbot accuracy tremendously, eliminating most of those agent requests.”

In one other case research, CTGT labored with an unnamed Fortune 10 firm to boost on-device AI capabilities in computationally constrained environments. The firm additionally helped a number one laptop imaginative and prescient agency obtain 10x sooner mannequin efficiency whereas sustaining comparable accuracy.

The firm claims its know-how can scale back hallucinations by 80-90% and allow AI deployments with 99.9% reliability, a crucial issue for enterprises in regulated industries like healthcare and finance.

From Hyderabad to Silicon Valley

Gorlla’s journey is itself outstanding. Born in Hyderabad, India, he mastered coding at age 11 and was disassembling laptops in highschool to squeeze out extra efficiency for coaching AI fashions. He got here to the United States to check on the University of California, San Diego, the place he acquired the Endowed Chair’s Fellowship.

His analysis there centered on understanding the elemental mechanisms of how neural networks be taught, which led to his ICLR paper and finally CTGT. In late 2024, Gorlla and co-founder Trevor Tuttle, an professional in hyperscalable ML programs, have been chosen for Y Combinator’s Fall 2024 batch.

The startup has attracted notable buyers past its institutional backers, together with Mark Cuban and different distinguished know-how leaders drawn to its imaginative and prescient of creating AI extra environment friendly and reliable.

Funding and future

Founded in mid-2024 by Gorlla and Tuttle, CTGT raised $7.2 million in February 2025 in an oversubscribed seed spherical led by Gradient, Google’s early-stage AI fund. Other buyers embody General Catalyst, Y Combinator, Liquid 2, Deepwater, and notable angels equivalent to François Chollet (creator of Keras), Michael Seibel (Y Combinator, co-founder of Twitch), and Paul Graham (Y Combinator).

“CTGT’s launch is timely as the industry struggles with how to scale AI within the current confines of computing limits,” stated Darian Shirazi, Managing Partner at Gradient. “CTGT removes those limits, enabling companies to rapidly scale their AI deployments and run advanced AI models on devices like smartphones. This technology is critical to the success of high-stakes AI deployments at large enterprises.”

With AI mannequin dimension outpacing Moore’s Law and advances in AI coaching chips, CTGT goals to deal with a extra foundational understanding of AI that may deal with each inefficiency and more and more complicated mannequin choices. The firm plans to make use of its seed funding to broaden its engineering workforce and refine its platform.

Each finalist offered to an viewers of 600 trade decision-makers and acquired suggestions from a panel of enterprise capital judges from Salesforce Ventures, Menlo Ventures, and Amex Ventures.

Read in regards to the different winners Catio and Solo.io. The different finalists have been Kumo, Superduper.io, Sutro and Qdrant.

Editor’s word: As a thank-you to our readers, we’ve opened up early hen registration for VB Transform 2026 — simply $200. This is the place AI ambition meets operational actuality, and also you’re going to need to be within the room. Reserve your spot now.