CTGT Earns Best Presentation Award at VB Transform 2025
▼ Summary
– CTGT, a startup focused on AI trustworthiness, won the Best Presentation Style award at VB Transform 2025 for its feature-level model customization technology.
– CTGT’s approach addresses the “AI Doom Loop,” where 54% of businesses cite AI as their highest tech risk, by offering faster, more accurate model training and customization.
– The company’s technology enables real-time modifications to AI models without retraining, with applications in compliance workflows and brand alignment tools.
– CTGT has demonstrated measurable ROI, including saving $5 million in liability for an insurer and reducing chatbot hallucinations for financial clients.
– Founded in 2024, CTGT raised $7.2 million in seed funding and aims to expand its team and refine its platform to address AI scalability challenges.
The AI industry faces growing challenges around trust and efficiency, but one startup is offering a fresh approach that’s turning heads. CTGT, a San Francisco-based company specializing in feature-level AI customization, recently took home the Best Presentation Style award at VB Transform 2025. Founded by 23-year-old Cyril Gorlla, the firm demonstrated how its unique technology helps enterprises overcome AI trust barriers by modifying models at the feature level rather than relying on traditional fine-tuning or prompt engineering.
Gorlla’s presentation tackled what he calls the “AI Doom Loop”, a cycle where businesses invest heavily in AI but struggle with unreliable outcomes. Citing industry reports, he noted that 54% of companies view AI as their top tech risk, while 44% have faced negative consequences from poorly implemented systems. “Many AI projects fail because there’s no fundamental trust in how these models behave,” Gorlla explained, referencing a recent case where a major corporation scrapped hundreds of AI pilots due to lack of ROI.
A Breakthrough in AI Customization
Traditional interpretability tools often require secondary AI systems for monitoring, but CTGT’s solution provides mathematically verifiable insights without extra computational overhead. By pinpointing specific neurons or feature directions responsible for issues like hallucinations or censorship, the technology dynamically adjusts model behavior without retraining. This allows enterprises to customize AI in real time, a game-changer for industries where reliability is non-negotiable.
Proven Enterprise Applications
“Companies no longer need to fine-tune hundreds of models for different use cases,” Gorlla emphasized. “Our system is model-agnostic, so they can plug it into existing workflows.”
One standout example involved DeepSeek models, where CTGT successfully modified features linked to censorship. The adjustments achieved a 100% response rate on sensitive queries while preserving performance in reasoning, math, and coding tasks.
As AI models grow more complex, CTGT’s focus on efficiency and trust positions it as a key player in the next wave of AI innovation.
(Source: VentureBeat)
