San Francisco-based CTGT, a startup targeted on making AI extra reliable by way of feature-level mannequin customization, gained the Greatest Presentation Type award at VB Rework 2025 in San Francisco. Based by 23-year-old Cyril Gorlla, the corporate showcased how its know-how helps enterprises overcome AI belief limitations by instantly modifying mannequin options as an alternative of utilizing conventional fine-tuning or immediate engineering strategies.
Throughout his presentation, Gorlla highlighted the “AI Doom Loop” confronted by many enterprises: 54% of companies cite AI as their highest tech threat based on Deloitte, whereas McKinsey stories 44% of organizations have skilled detrimental 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 major departure from standard AI customization methods. The corporate was based on analysis Gorlla performed whereas holding an endowed chair on the College of California San Diego.
In 2023, Gorlla revealed a paper on the Worldwide Convention on Studying 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.
Relatively than counting on brute-force scaling or conventional deep studying strategies, CTGT has developed what it calls an “entirely new AI stack” that basically reimagines how neural networks study. The corporate’s innovation focuses on understanding and intervening on the function stage of AI fashions.
The corporate’s method differs basically from customary interpretability options that depend on secondary AI methods for monitoring. As a substitute, CTGT gives mathematically verifiable interpretability capabilities that eradicate the necessity for supplemental fashions, considerably reducing 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 habits on the fly with out taking methods offline for retraining.
Actual-world purposes
Throughout his Rework presentation, Gorlla demonstrated two enterprise purposes already deployed at a Fortune 20 monetary establishment:
A model alignment device that helps entrepreneurs develop copy in keeping with model values. The system can counsel personalised recommendation on why sure phrases work effectively for a selected model and the best 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 answerable for censorship behaviors. By isolating and adjusting these particular activation patterns, CTGT was capable of obtain a 100% response price on delicate queries with out degrading the mannequin’s efficiency on impartial duties like reasoning, arithmetic and coding.
Photographs: CTGT presentation at VB Rework 2025
Demonstrated ROI
CTGT’s know-how seems to be delivering measurable outcomes. Through 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.”
One other early buyer, Ebrada Monetary, 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,” mentioned 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 reinforce on-device AI capabilities in computationally constrained environments. The corporate additionally helped a number one laptop imaginative and prescient agency obtain 10x sooner mannequin efficiency whereas sustaining comparable accuracy.
The corporate 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 US to check on the College of California, San Diego, the place he acquired the Endowed Chair’s Fellowship.
His analysis there targeted on understanding the elemental mechanisms of how neural networks study, which led to his ICLR paper and ultimately CTGT. In late 2024, Gorlla and co-founder Trevor Tuttle, an skilled in hyperscalable ML methods, 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 outstanding know-how leaders drawn to its imaginative and prescient of creating AI extra environment friendly and reliable.
Funding and future
Based 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. Different buyers embody Common Catalyst, Y Combinator, Liquid 2, Deepwater, and notable angels comparable 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,” mentioned Darian Shirazi, Managing Associate 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 Legislation and advances in AI coaching chips, CTGT goals to give attention to a extra foundational understanding of AI that may address each inefficiency and more and more complicated mannequin choices. The corporate plans to make use of its seed funding to increase its engineering crew and refine its platform.
Every 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.
Learn in regards to the different winners Catio and Solo.io. The opposite finalists have been Kumo, Superduper.io, Sutro and Qdrant.
Editor’s observe: As a thank-you to our readers, we’ve opened up early fowl registration for VB Rework 2026 — simply $200. That is the place AI ambition meets operational actuality, and also you’re going to wish to be within the room. Reserve your spot now.
Every day insights on enterprise use instances with VB Every day
If you wish to impress your boss, VB Every day has you coated. We provide the inside scoop on what firms are doing with generative AI, from regulatory shifts to sensible deployments, so you possibly can share insights for max ROI.
An error occured.
