SpeechTech entrepreneur Steve Young takes chair at Inephany

Amadeus Capital Partners, based in Cambridge and London, this week led a pre-seed investment of $2.2 million in Inephany. Amadeus was joined in the funding round by Sure Valley Ventures and Professor Young, who is a venture partner at the VC firm.

Founded in July 2024 by Dr John Torr (formerly of Apple Siri’s machine learning team), Hami Bahraynian, and Maurice von Sturm (co-founders of conversational AI startup Wluper), Inephany brings together deep technical expertise in neural network optimisation. The funding will be used to grow the core engineering team, advance its optimisation platform, and onboard its first enterprise customers.

Professor Young, renowned for his foundational contributions to speech recognition and dialogue systems -including key work behind Apple’s Siri - has joined as chair to help guide the company’s next phase of growth.

Inephany uses AI to optimise training in real time, cutting compute costs and accelerating development. Its AI-powered optimisation platform promises to revolutionise how neural networks - including Large Language Models (LLMs) - are trained and fine-tuned.

As generative AI continues its rapid ascent, the soaring compute and energy costs of training cutting-edge models have emerged as a major bottleneck.

Training GPT-4 is estimated to have cost between $60 million and $100m, with next-generation models edging towards the $1 billion mark according to industry leaders such as Anthropic.

AI compute demands are now doubling roughly every six months, outpacing Moore’s Law and rendering traditional training and optimisation methods increasingly unsustainable. Inephany addresses this challenge head-on with a novel AI-driven optimisation system that intelligently controls the training process in real-time.

Compared to traditional brute force approaches that rely on exhaustive trial-and-error optimisations, Inephany’s technology dramatically improves sample efficiency, accelerates training, reduces overall development time and enhances final model performance—all while slashing compute costs.

This breakthrough holds the potential to unlock scalable, sustainable AI development that’s at least 10x more cost-effective.

While the company’s initial focus is on training-time optimisation for LLMs, its technology has broad applicability across the AI landscape—from Recurrent Neural Networks used in financial time-series forecasting, to Convolutional Neural Networks powering computer vision in autonomous systems.

The company also has plans to expand its AI-powered optimisation approach to inference time compute. By slashing the cost and compute burden of training and deploying these models, the Inephany team aims to democratise access to advanced AI and accelerate innovation across industries.

John Torr, CEO at Inephany, said: “We are thrilled to be backed by such experienced investors, and having a seasoned entrepreneur and AI pioneer like Professor Steve Young as our chair is a true privilege.

“Current approaches to training LLMs and other neural networks are extremely wasteful across multiple dimensions. Our unique solution tackles this inefficiency head-on, with the potential to radically reduce both the cost and time required to train and optimise state-of-the-art models.

“As we prepare to deliver our first products later this year, we are incredibly excited to embark on the next chapter of our journey—and to help shape the ongoing AI revolution by transforming AI optimisation.”

Amelia Armour, Partner at Amadeus Capital Partners said: “We very much look forward to backing John, Hami, and Maurice as they tackle key efficiency challenges in current AI training. Their innovative approach to automating and optimising neural network training has the potential to reduce costs by an order of magnitude and accelerate advancements across AI applications. If rolled out at scale, the impact of this on what models can deliver will be very substantial.”

Professor Young, added: “As the use of AI spreads ever wider, moving beyond the traditional applications of speech, language and vision into new and diverse areas such as weather prediction, healthcare, drug discovery and materials design, the need for very efficient training of accurate neural models is becoming critical.

“The groundbreaking new approach being developed by Inephany marks a step change in neural model training technology and I am delighted to join the team as chair and investor.”