In the world of asset management, where information flows like a torrent and market events demand swift responses, the ability to rapidly deploy custom AI-powered applications has become paramount. Infant Vasanth, Senior Director of Engineering, and Vaibhav Page, Principal Engineer at BlackRock, recently spoke at the AI Engineer World's Fair, illuminating their innovative approach to scaling custom AI knowledge applications. Their discussion highlighted the critical role of these tools in synthesizing vast amounts of data, developing investment strategies, and rebalancing portfolios, all while navigating a heavily regulated industry.
The sheer volume of information and the complexity of financial instruments necessitate a constant stream of bespoke internal tools. Building these applications traditionally is a protracted affair, often taking "somewhere between three to eight months to build a single app for a complex use case," Vasanth explained. Challenges span prompt engineering, where prompts can quickly balloon from a few sentences to "three paragraphs long," requiring meticulous management, iteration, and quality evaluation. Furthermore, Large Language Model (LLM) strategies contend with issues like bad retrieval, hallucination, and inherent context limitations, especially when dealing with documents "thousands of pages long." Finally, the deployment of these applications introduces its own set of hurdles, from domain federation and access control to resource management, model latency, and cost.
To overcome these obstacles, BlackRock developed DocuFlow, a modular, Kubernetes-native AI framework designed to drastically accelerate the development and deployment of these critical knowledge apps. This innovative architecture focuses on empowering domain experts and streamlining the entire lifecycle from concept to production.
At its core, DocuFlow features a "Sandbox" environment. This playground allows operators and subject matter experts (SMEs) to quickly build and refine extraction templates, run extractions on document sets, and compare results. This shift in ownership is pivotal, as Vasanth emphasized, "if you are able to federate out those pain points or those bottlenecks... into the hands of the domain experts, then your iteration speed becomes really fast."
The framework also includes an "App Factory," described as a cloud-native operator that takes the refined definitions from the Sandbox and automatically spins up functional applications. This significantly reduces the time from ideation to deployment. The orchestration layer, powered by the DocuFlow API, seamlessly manages embeddings, various extraction workflows, transformation workflows, and execution, ensuring smooth data flow and processing. This holistic approach has enabled BlackRock to compress app development timelines from months to mere days, unlocking new levels of automation and efficiency across investment management workflows.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 64,000 other subscribers.
A crucial insight from BlackRock's experience, particularly relevant in a regulated financial environment, is the indispensable role of human oversight. As Vasanth pointed out, "Human in the loop is super important... in the financial space with compliance, with regulations, you kind of need those four eyes check and you kind of need the human in the loop. So design for human in the loop first if you are in a highly regulated environment." This ensures accuracy, mitigates risks like hallucination, and maintains compliance, acknowledging that full automation isn't always the optimal or safest path. By embedding human review and validation at key stages, BlackRock can leverage AI's speed while upholding the rigorous standards of the finance industry.
