Generative AI (GenAI), encompassing Large Language Models (LLMs) and other modalities like image generation, is transforming industries by improving efficiency and productivity. This versatile technology can read, write, code, and draw, enhancing creativity and productivity across various sectors including retail, finance, manufacturing, and healthcare. When used strategically, GenAI can drive innovation and offer a competitive edge in today's data-driven landscape.
However, building enterprise-grade GenAI solutions comes with challenges:
Trust and Safety: Ensuring data privacy, ethical use, and security safeguards.
Cost and Scale: Managing costs influenced by Large Language Model (LLM) size and capabilities, handling latency issues, and balancing demand amid rationed quotas and GPU shortages.
Changing Landscape: Keeping pace with rapid innovation without compromising business goals.
CB-GPT, short for Coinbase-GPT, serves as a unified platform for all GenAI applications at Coinbase. By building and managing our own GenAI platform, we are able to tailor our solutions to unique crypto use cases and Coinbase customer segments. CB-GPT has a two-fold Vision:
Source: Coinbase
Source: Coinbase
As noted above, three key challenges for GenAI deployment include trust and safety, cost and scale, and changing landscape. Here’s how CB-GPT addresses these challenges:
CB-GPT integrates cloud-hosted LLMs from various providers like Azure, GCP, and Anthropic, as well as open-source options such as LLaMa, and Mistral that are self-hosted on Coinbase infrastructure. It intelligently selects LLMs to optimize cost and scale for specific tasks, adapting to the evolving landscape of LLM technology, while maintaining a consistent interface. This flexibility allows applications to transition to newer and better technologies, regardless of their source.
CB-GPT provides access to both enterprise and web data, as appropriate, ensuring that LLM RAG (Retrieval Augmented Generative) responses are “grounded” in trusted sources of truth, while maintaining stringent access controls.
CB-GPT incorporates security and privacy guardrails, along with access control mechanisms, to facilitate safe and secure GenAI application development. It also incorporates frameworks for evaluating LLM tasks.
Further, CB-GPT supports a diverse set of technical and non-technical users, by making its features accessible both via an API and a no-code UI tool, allowing staff from diverse job functions to build out relevant solutions rapidly.
With CB-GPT, our goal is to deliver intuitive experiences through our products, offering personalized and contextual support at any time. We aim to create rich opportunities for our wide range of users (from individuals to institutional investors) to learn, explore, and participate in the crypto economy.
Coinbase's CX Retail Ops and CX Compliance Ops face significant workload challenges. CX Retail Ops must educate customers on crypto and Coinbase products, addressing financial literacy and technostress due to complex crypto concepts. Meanwhile, CX Compliance Ops is tasked with adhering to evolving regulations across 100+ geographies, involving laborious processes like transaction monitoring, user screening and risk assessment. Leveraging CB-GPT offers a substantial opportunity to enhance efficiency by assisting these critical functions.
CX Agent Assist: Enable agents to utilize CB-GPT to assist with case investigation and translations.
CX Compliance Assist: Improve agent productivity in Compliance, combing through thousands of pages of documented regulatory requirements, and drafting summaries for agent review to expedite reporting.
We believe in democratizing the power of LLMs for use across the entire employee base of the company.
Employee Assistant: Designed to help Coinbase employees with routine tasks including drafting documents, updating project status, providing quick responses to FAQs on Slack, summarizing documents, and rephrasing textual content. Available as a Web UI, Slack app, and Chrome extension
CB-GPT Studio: Self-serve no-code UI tool enabling Coinbase employees to create and maintain AI assistants for bespoke tasks.
AI Assistants for Specific Job Functions: Several other job functions within Coinbase utilize proprietary or third-party tools to enhance their work. CB-GPT provides seamless integrations with these tools, further boosting productivity for these professionals.
We are committed to bringing the power of GenAI into the daily workflows of Coinbase employees and customers. Our endeavor is to move forward swiftly, yet cautiously, ensuring that all necessary safeguards are in place in this fast-paced innovative landscape.
About Rajarshi Gupta, Katie McCann and Varsha Mahadevan
Rajarshi Gupta is the Head of Machine Learning at Coinbase, bringing smart automation and protection to crypto users around the world. Prior to this, Rajarshi was GM, ML Services at AWS. He also worked for many years at Qualcomm Research, where he created ‘Smart Protect’, the first ever product to achieve On-Device Machine Learning for Security and shipped in over 1 billion Snapdragon chipsets. Rajarshi has a PhD in EECS from UC Berkeley and has built a unique expertise at the intersection of Artificial Intelligence and Blockchains. Rajarshi is a prolific inventor and has authored 225+ issued U.S. Patents.
Katie McCann currently leads the Machine Learning product team at Coinbase, where she spearheads the development of AI-driven solutions across numerous domains including Risk, Recommendations, and Customer Support. Before joining Coinbase, Katie spent several years at Meta, where she focused on developing Machine Learning models to improve the quality and relevance of what users see on Facebook. With a wealth of experience from startups to big tech, Katie blends her technical expertise with strategic vision to drive impactful solutions.
Varsha Mahadevan is the Senior Engineering Manager leading the CB-GPT team at Coinbase, where she drives the integration of Generative AI into the everyday workflows of employees and customers. Before joining Coinbase, Varsha served as the Associate Vice President of Engineering at BankBazaar.com, a leading fintech company in India. She brings a wealth of experience from her tenure at Microsoft, where she played a pivotal role in developing the .Net Framework and creating AI-driven personalization features for Cortana, Microsoft’s AI assistant.