Meet Peng Qi, Senior Director of AI Science

Every day, Uniphore’s AI Science Team comes together with one goal in mind: how to solve real-world challenges with AI. It’s a passion that connects everyone on our staff—from our AI development leaders and distinguished scientists to the talented engineers that bring our innovations to life. And, for most, it began long before joining our industry-leading team.

For Peng Qi, Senior Director of AI Science, that passion took root as a computer software undergraduate at Tsinghua University in Beijing and, later, as a doctorate student at Stanford University. In a recent interview, he credited the meteoric rise of artificial intelligence—and its potential to impact positive change—as his main motivation for pursuing a career in AI. “I was fortunate to graduate into a world where the opportunities are ripe for AI to shine in real-world applications.”

After acquiring his PhD in Computer Science, Peng jumped at the chance to lead that exploration, first as a Senior Research Scientist for JD.com—China’s largest retailer by revenue—and then as a Senior Applied Scientist for Amazon Web Services.

Both companies, he explained, have no shortage of challenging problems with real-world considerations. Due to their scale, AI solutions must be efficient and reliable before they are productionized. As a result, their development requires a systematic approach rather than one that focuses solely on AI models.

“For instance, to deliver shorter latency to customers when building enterprise chat assistants at Amazon Q, we had to not only evaluate options of various AI models, but also deep dive into engineering solutions to parallelize prerequisites to the final response where possible,” he said. “At the same time, we explored product design solutions of showing the user some intermediate system operations to promote transparency.” 

Peng’s time with AWS also taught him to work backwards from problems in category-defining products. For example, when tasked with generating accurate citations in chat assistant responses in Amazon Q, Peng’s team developed an innovative approach called CiteEval that leveraged LLMs as a judge to systematically understand the various sources of knowledge in the assistant response. A pioneering study, CiteEval would form the basis of many future innovations in the field.

Describing his takeaway from the experience, Peng said, “Product decisions should be driven by vetted quantitative studies to reflect customer experience once the product is deployed. This requires cross-functional alignment on how progress should be measured and decisions made in the development of AI products.” 

That strategy—as well as his extensive experience with LLMs, LAMs, neuro-symbolic and SOTA digital agents—would later guide him as Head of Research at Orby AI before joining Uniphore as Senior Director of AI Science. (Orby AI was acquired by Uniphore in 2025.) 

In his current role, Peng leads both frontier research and product development of AI tools that aim to help non-technical or less-technical teams build reliable AI tools for themselves.  

To achieve this goal, Peng plugs himself and his team at both the bleeding edge of AI technology and the frontier of product design and customer needs. “Besides being nimble and reactive to technology and market trends, we constantly make predictions about AI technology and product trends in the coming months and revisit them to calibrate our understanding before making new ones regularly,” he explained. “This allows us to stay ahead of the curve and avoid investment in technology and product that will be too early to market or behind the curve in a saturated market once we finish building it.” 

This approach keeps Peng’s team focused on solving the right challenges that would best serve Uniphore’s customers down the line. Chief among those challenges is enabling customers without deep AI knowledge to develop advanced AI agents and, consequently, unlock the full value of AI on their own. To solve this challenge—and answer this growing customer need—his team is exploring agentic solutions that transcend the current limitations of LLMs. 

“With LLM capabilities rapidly improving in the past years, AI agents have entered into the enterprise workplace as a potential solution to reducing the amount of tedium and repetitiveness in many people’s daily routine,” he said. “However, powerful as LLMs are, it typically requires non-trivial work to set up accurate and reliable AI agents for enterprise tasks of meaningful complexity, and the cost of running such agents with LLMs will often remain prohibitively high for them to be economically viable.”  

The solution: SLMs powered by the technology behind large action models. Using this innovative formula, Peng and his team are empowering customers to easily instrument agentic workflows at scale—including use cases where data sovereignty prohibits using such data for LLM training or inference—and drive significant ROI as a result. 

But that’s not all. Peng revealed that customers will soon be able to finish the entire SLM building lifecycle within the Uniphore Business AI Cloud—without needing to configure complex AI infrastructure or gaining deep knowledge about SLM training. 

“I’m excited about the reimagined SLM finetuning studio experience that we are currently building,” he shared. “With an agentic tool, customers will get step-by-step guidance in their journey to build SLMs for enterprise use cases they care about and seamlessly deploy these SLMs in their production environment.” 

Like other innovations his team has led, this solution is grounded in real-world need. That’s an important distinction, Peng stresses. With AI technologies making inroads in so many aspects of our lives, he encourages AI practitioners to take a step back and learn more about the world—and the problems yet be addressed—rather than focusing purely on innovation for innovation’s sake.

Peng’s advice for prioritizing problem-solving above all else isn’t just a best practice for AI scientists. It’s a guiding principle that sets Uniphore apart, aligning innovation with real customer needs.