Mainframe modernization to AI apps: USPTO reinvents itself
How the US Patent and Trade Office, founded in 1802, maintained IT resilience while managing systems ranging from mainframes to the cloud, AI and beyond.
The U.S. Patent and Trade Office knows plenty about innovation -- registering and tracking new inventions has been the agency's primary mission for more than 220 years. But five years ago, it realized it had a new challenge: revamping the internal IT systems it had accrued over the last six decades.
The initial moment of clarity came for the USPTO during an outage of the agency's homegrown Patent Application Locating and Monitoring (PALM) database from August 15 to August 23, 2018. PALM supported other USPTO systems for filing, searching and paying patent fees. During the outage, patent applicants had to use other ways to file, and the USPTO was forced to issue refunds for paper filing fees applicants incurred. The outage made it clear that IT practices must change.
Two years after, CIO Jamie Holcombe was hired to further expand the organization's digital transformation efforts. He hired CTO Stephan Mitchev and other new IT leaders and set about instilling a broad swath of technical and organizational changes.
The agency's progress since then, while significant, has also been gradual, Holcombe said.
"It took about 18 to 22 months before people started seeing some real results," he said. "But once they see it, then the momentum is built. Now I've got to spread that it into the larger business, and that's where I'm at right now at the four-year mark."
Mainframe modernization dislodges ALGOL
The latest milestone for USPTO's database refresh came last month in its Trademark Distribution Information System. USPTO successfully decoupled that system from a forty-year-old mainframe application, the Trademark Register Application Monitoring (TRAM) system, which was written in Algorithmic Language (ALGOL), a predecessor language to COBOL that dates back to the 1960s.
Jamie Holcombe
Three USPTO staffers who are still familiar with the ALGOL system helped the rest of the development team translate TRAM into a distributed computing app compatible with Oracle databases. TRAM contains more than 800 individual services -- and the complications don't end there, according to Holcombe.
"The old TRAM is actually synchronized with a new CRM system, so we've been having to update both in order to make sure there's good data quality and integrity," he said. "That's our big focus this year, and then we can finally take that anchor away from our neck."
Once that work is complete in September, Holcombe predicted, database development teams will be able to move on to new ways of working -- namely Agile and DevOps methodologies accompanied by infrastructure automation via Kubernetes and containers. For now, IT teams in charge of these database apps make manual updates to them on servers in an Iron Mountain collocation data center, with some infrastructure and test automation via Puppet and Jenkins, respectively.
DevSecOps and the product mindset
The mainframe modernization team is an example of what occurred on the human side of USPTO's IT transformation. The agency added a set of about 220 equal but decoupled groups within its traditional hierarchical organizational chart. Vestiges of the hierarchical system remain where they are necessary, Holcombe said. But within it, employees from various departments are assigned for a short period of time to a smaller group.
"It's a team that actually can do everything it needs to do for that mission," he said. "The [hierarchy] becomes responsible for ensuring that all those different teams are resourced."
Spence Spencer
Those groups then take ownership of the full lifecycle of each of the agency's IT services, operating them like products for their entire lifecycle rather than as a short-term project. This approach, also known as a product mindset, has allowed USPTO to maintain IT resilience amid all its other technical transitions, said Spence Spencer, director of the system configuration and delivery automation division in the USPTO's Office of the CIO.
For example, in February, Spencer's team responded to a critical security vulnerability that required redeployment of an application component. Part of the total lead time of approximately 19 hours to correct the vulnerability was waiting for a window outside regular business hours to update production, but the update was deployed to a test environment within hours, Spence said.
"So a government agency, in under 24 hours, did the entire software lifecycle, from conception all the way through to deployment," Spence said.
In the past, such changes could take up to six months, according to Holcombe.
In addition to making fixes faster, the product mentality among product teams encourages them to build resilience into their services from the beginning, Spencer said, including security features.
"That team knows that they're going to be living with this thing in five years," he said. "They have a strong incentive to get it right."
For new apps, USPTO teams take a DevSecOps approach using features such as static code analysis and vulnerability detection built into the GitLab CI/CD pipeline that automates deployments to AWS cloud infrastructure. This keeps some vulnerable code from reaching production in the first place, Spencer said.
Tensorflow, ChatGPT lead AI apps efforts
Mainframe modernization and DevSecOps remain works in progress, but USPTO has also jumped into its next phase of digital transformation: automation through artificial intelligence. A group of a dozen engineers from USPTO earned certifications in Google's TensorFlow open source machine learning platform two years ago. That investment has already yielded results, Holcombe said.
That team knows that they're going to be living with this thing in five years. They have a strong incentive to get it right.
Spence SpencerDirector of system configuration and delivery automation division, USPTO
"We have to classify our patents when they come in, and we're using contractors to [do it]," Holcombe said. "But we've also been having the neural network learn classification. And it's gotten to the point where we're able to release over half of the humans that were doing it -- that's tens of millions of dollars."
Next, USPTO will investigate how to use ChatGPT to augment examiners doing patent searches for prior art when evaluating new applications. Tests so far have revealed that the ChatGPT algorithm can personalize results for each examiner, according to Holcombe.
"Each examiner is always a unique entity, and the AI is giving them what they want," he said. "It's remembering their choices and the relative nature of what they think is important."
Beth Pariseau, senior news writer at TechTarget, is an award-winning veteran of IT journalism. She can be reached at [email protected] or on Twitter @PariseauTT.
