Executive summary

Qualcomm, a global leader in the semiconductor industry, is renowned for its pioneering work at the intersection of mobile connectivity and artificial intelligence. With a bold vision to create a more connected and intelligent world, Qualcomm Technologies’ relentless pursuit of AI-driven solutions is reshaping today’s technology and laying the foundation for a smarter, more integrated future. Qualcomm Technologies Vice President of Engineering, Yogi Chiniga, and his team sought a next-generation AI solution to enhance and scale their highly specialized engineering workforce. After evaluating several RAG solutions and finding them to be inadequate, the company selected Contextual AI to help transition beyond the demo phase towards full-scale production. As of November 2024, the resulting solution has been smoothly incorporated into the daily workflows of thousands of engineers across various technology areas, resulting in increased efficiency and enhanced productivity.

The Production Challenge

The Customer Engineering organization at Qualcomm Technologies plays a vital role in supporting thousands of customers, including major companies, with all aspects of product development—from design and development to debugging and integration. The team tackles a broad array of complex technical challenges, requiring deep expertise across multiple Qualcomm® technology areas. However, like many teams with highly specialized technical talent, Customer Engineering faced increasing bandwidth constraints. To address the growing demands on the team, Qualcomm Technologies leadership sought an AI solution that would help maximize the team’s productivity and deliver an excellent customer experience. They envisioned an application capable of efficiently querying and synthesizing the company’s vast technical documentation, both internal and external-facing, ultimately reducing the time needed to resolve certain customer issues.

The project’s high accuracy requirements and low tolerance for failure made it exceptionally challenging for standard RAG (retrieval-augmented generation) solutions to deliver the accuracy needed to be considered production-ready:

1. Massive volumes of noisy data: For most RAG solutions, the ability to retrieve the right information degrades as the volume of data increases, especially in real-world deployments where data quality can be inconsistent. This made Qualcomm Technologies’ use case, which required handling millions of pages of highly technical documents, with thousands of new pages added each day, particularly difficult to implement.
2. Complex, multimodal content: The company’s documentation included complex diagrams, technical images, nested tables, and code, spread across a diverse set of data formats (PDF, HTML, Excel, and more), complicating extraction and accurate retrieval.
   Documentation includes complex figures, specialized concepts, and technical terminology that make extraction and accurately retrieval challenging
3. Diversity of domains: The solution needed to span a large number of fundamentally distinct technology areas (e.g., Multimedia, Modem, Hardware, RF), each with unique terminology, documentation conventions, and internal acronyms.

Achieving the “production bar” with Contextual AI

To achieve the level of accuracy required for a production deployment, Qualcomm Technologies collaborated with Contextual AI, the leader in production-grade RAG. Unlike most alternative solutions, which struggled to deliver the required precision, Contextual AI’s platform—fueled by their breakthrough RAG 2.0 approach—optimized the RAG pipeline as a unified system and helped specialize the system for the Customer Engineering use case. The solution enabled Qualcomm Technologies to answer highly specialized questions with production-level accuracy, and paved a clear path to a global production deployment across thousands of customer engineers.

Contextual AI’s platform enabled Qualcomm Technologies to deploy highly accurate and specialized RAG applications to its customer engineers

Delivering “ROI of AI” for Qualcomm Technologies’ Customer Engineering

Contextual AI’s specialized applications achieved high accuracy across a wide variety of deeply technical questions and documents, while providing fine-grained attributions in the source documentation

Many companies struggle with moving beyond the “demo stage” of AI to a real-world, high-value production setting, especially if the underlying data sources are vast and complex, as was the case with Qualcomm Technologies. Thanks to Contextual AI, Qualcomm Technologies was able to achieve:

1. High accuracy at scale: The solution exceeded production-grade accuracy thresholds across all technology areas. With the RAG components jointly optimized as a single system and further specialized using the platform’s fine-tuning and alignment engine, Qualcomm Technologies maintained production-level accuracy even as the number of users and the volume of underlying data increased dramatically.
   
2. Rapid deployment to production: Developing high-quality RAG applications is a complex undertaking, and standard RAG techniques were not up to par for the complexity and criticality of this use case. Thanks to the hands-on collaboration with Contextual AI and its cutting edge technology, Qualcomm Technologies was able to move to production quickly, enabling customer engineers to use the solution to increase their efficiency and productivity.
3. Broad internal adoption: Qualcomm Technologies embedded its application directly into existing systems for handling customer requests, expediting user adoption and seamlessly integrating with the existing workflows for customer engineers. To date, thousands of end users have been onboarded to the application.

What’s next for Qualcomm Technologies and Contextual AI

While Qualcomm Technologies’ first Contextual AI application was designed for internal usage, the team plans to provide a similar application directly to customers soon. By enabling customers to use the Contextual AI solution to directly interact with Qualcomm’s rich and complex documentation, the Customer Engineering organization expects to significantly reduce the volume of documentation-related support. Qualcomm Technologies and Contextual AI are also actively exploring agentic RAG workflows, formulating initial search queries automatically across diverse data sources.

Beyond Customer Engineering, Qualcomm Technologies is considering using the Contextual AI Platform for a variety of additional complex and high-value use cases that require clearing challenging accuracy thresholds with the type of specialization that only a RAG 2.0 approach can provide.

Qualcomm branded products are products of Qualcomm Technologies, Inc. and/or its subsidiaries.