Last week we announced the private preview of Duet AI for Google Cloud, an always-on AI collaborator that uses generative AI to provide help to developers and cloud users. This article gives you a detailed look at Duet AI for developers, showing how Duet AI can help provide developers with real-time code suggestions, chat assistance, and enterprise-focused customization. You can sign-up here to join our waitlist for Google Cloud’s AI trusted tester program.  

We believe that addressing these use case​​s with large language models (LLMs) will usher in a major leap in productivity in enterprise development. Duet AI uses Codey, a family of code models built on PaLM 2. 

Developers continuously seek ways to improve their productivity and, over the past few decades, these efforts have resulted in massive productivity leaps due to technological changes. From advanced debuggers and online developer communities, to modern IDEs/notebooks and cloud computing, each advance brought massive changes in productivity. Despite these improvements, developers still face numerous challenges, some of which are unique to cloud development:

• Disruptive context switching and friction when integrating a new tool or service
• Excessive time spent on repetitive tasks
• Time required to understand a new code base or project
• Large cognitive workload when working on large code bases or complex APIs

Duet AI for developers focuses on challenges and tasks across the development lifecycle:

Code/Boilerplate Generation — Developers can describe the tasks they have in mind as a comment or function name, such as creating a Cloud Pub/Sub topic. Duet AI will generate a reference implementation that can be reviewed and modified, so developers don’t need to spend time reading through multiple documentation pages.

Inline Code Completion — To reduce the time spent on repetitive tasks and minimize the cognitive workload of tasks such as writing repetitive code or retrieving variable names, Duet AI provides intelligent, context-aware code completion, helping reduce the time spent on coding and enhancing the quality of the written code.

Enterprise Customization — Organizations frequently have massive code bases and specific recommended frameworks and best practices, which generic code assistance solutions may not be best positioned to support. With Vertex AI, developers will be able to tune and customize the underlying models and connect them to the Duet AI experience, allowing for assistance optimized to the needs of the organization.

Code Explanation — Developers spend significant time and effort reading and understanding code written by their peers or external contributors. To help assist in this process, Duet AI for code assistance provides an “Explain this code” option available whenever a developer selects their code, allowing them to more quickly understand, map, and navigate unfamiliar code bases.

Code security guardrails — Code generated by Duet AI can also be scanned for vulnerable dependencies via Source Protect, helping surface known public vulnerabilities impacting code, along with suggested fixes when available, bringing additional security.

By harnessing the power of AI-driven developer assistance such as the one provided by Duet AI for developers, businesses can unlock unprecedented levels of productivity and efficiency in software development, paving the way for a new era of innovation and growth.

These early features of Duet AI for Google Cloud will be available for limited users and we will be expanding access very soon. Sign up here to join Google Cloud’s AI Trusted Tester Program.

Digitec Galaxus AG is the biggest online retailer in Switzerland, operating two online stores: Digitec, Switzerland’s online market leader for consumer electronics and media products, and Galaxus, the largest Swiss online shop with a steadily growing range of consistently low-priced products for almost all daily needs. 

Known for its efficient, personalized shopping experiences, it’s clear that Digitec Galaxus understands what it takes to deliver a platform that is interesting and relevant to customers every time they shop. 

The problem: Personalizing decisions for every situation

Digitec Galaxus already had established an engine to help them personalize experiences for shoppers when they reached out to Google Cloud. They had multiple recommendation systems in place and were also extensive early adopters of Recommendations AI, which already enabled them to offer personalized content in places like their homepages, product detail pages, and their newsletter. 

But those same systems sometimes made it difficult to understand how best to combine and optimize to create the most personalized experiences for their shoppers. Their requirements were threefold:

1. Personalization: They have over 12 recommenders they can display on the app, however they would like to contextualize this and choose different recommenders (which in turn select the items) for different users. Furthermore they would like to exploit existing trends as well as experiment with new ones.
2. Latency: They would like to ensure that the solution is architected so that the ranked list of recommenders can be retrieved with sub 50 ms latency.
3. End-to-end easy to maintain & generalizable/modular architecture: Digitec wanted the solution to be architected using an easy to maintain, open source stack, complete with all MLops capabilities required to train and use contextual bandits models. It was also important to them that it is built in a modular fashion such that it can be adapted easily to other use cases which have in mind such as recommendations on the homepage, Smartags and more . 

To improve, they asked us to help them implement a machine learning (ML) contextual bandit based recommender system on Google Cloud taking all the above factors into consideration to take their personalization to the next level. 

Contextual bandits algorithms are a simplified form of reinforcement learning and help aid real-world decision making by factoring in additional information about the visitor (context) to help learn what is most engaging for each individual. They also excel at exploiting trends which work well, as well as exploring new untested trends which can yield potentially even better results. For instance, imagine that you are personalizing a homepage image where you could show a comfy living room couch or pet supplies. 

Without a contextual bandit algorithm, one of these images would be shown to someone at random without considering information you may have observed about them during previous visits. Contextual bandits enable businesses to consider outside context, such as previously visited pages or other purchases, and then observe the final outcome (a click on the image) to help determine what works best. 

Creating a personalization system with contextual bandits

While Digitec Galaxus heavily personalizes their website homepages, they are very very sensitive and also require more cross-team collaboration to update and make changes. 

Together with the Digitec Galaxus team, we decided to narrow the scope and focus on building a contextual bandit personalization system for the newsletter first. The digitec Galaxus team has complete control over newsletter decisions and testing various ML experiments on a newsletter would have less chance of adverse revenue impact than a website homepage. 

The main goal was to architect a system that could be easily ported over to the homepage and other services offered by Digitec with minimal adaptations. It would also need to satisfy the functional and non-functional requirements of the homepage as well as other internal use cases.

Below is a diagram of how the newsletter’s personalization recommendation system works:

• The system is given some context features about the newsletter subscriber such as their purchase history and demographics. Features are sometimes referred to as variables or attributes, and can vary widely depending on what data is being analyzed. 
• The contextual bandit model trains recommendations using those context features and 12 available recommenders (potential actions). 
• The model then calculates which action is most likely to enhance the chance of reward (a user clicking in the newsletter) and also minimize the problem (an unsubscribe). 

Calculating whether a click was a newsletter or an unsubscribe enabled the system to optimize for increasing clicks and avoid showing non-relevant content to the user (click-bait). This enabled Digitec Galaxus to exploit popular trends while also exploring potentially better-performing trends. 

How Google Cloud helps

The newsletter context-driven personalization system was built on Google Cloud architecture using the ML recommendation training and prediction solutions available within our ecosystem. 

Below is a diagram of the high-level architecture used:

The architecture covers three phases of generating context-driven ML predictions, including: 

ML Development: Designing and building the ML models and pipeline 
Vertex Notebooks are used as data science environments for experimentation and prototyping. Notebooks are also used to implement model training, scoring components, and pipelines. The source code is version controlled in Github. A continuous integration (CI) pipeline is set up to automatically run unit tests, build pipeline components, and store the container images to Cloud Container Registry. 

ML Training: Large-scale training and storing of ML models 
The training pipeline is executed on Vertex Pipelines. In essence, the pipeline trains the model using new training data extracted from BigQuery and produces a trained, validated contextual bandit model stored in the model registry. In our system, the model registry is a curated Cloud Storage. 

The training pipeline uses Dataflow for large scale data extraction, validation, processing, and model evaluation, and Vertex Training for large-scale distributed training of the model. AI Platform Pipelines also stores artifacts, the output of training models, produced by the various pipeline steps to Cloud Storage. Information about these artifacts are then stored in an ML metadata database in Cloud SQL. To learn more about how to build a Continuous Training Pipeline, read the documentation guide.

ML Serving: Deploying new algorithms and experiments in production 
The training pipeline uses batch prediction to generate many predictions at once using AI Platform Pipelines, allowing Digitec Galaxus to score large data sets. Once the predictions are produced, they are stored in Cloud Datastore for consumption. The pipeline uses the most recent contextual bandit model in the model registry to evaluate the inference dataset in BigQuery and give a ranked list of the best newsletters for each user, and persist it in Datastore. A Cloud Function is provided as a REST/HTTP endpoint to retrieve the precomputed predictions from Datastore.

All components of the code and architecture are modular and easy to use, which means they can be adapted and tweaked to several other use cases within the company as well.

Better newsletter predictions for millions

The newsletter prediction system was first deployed in production in February, and Digitec Galaxus has been using it to personalize over 2 million newsletters a week for subscribers. The results have been impressive, 50% higher than our baseline. However, the collaboration is still ongoing to improve the results even more. 

“Working at this level in direct exchange with Google’s machine learning experts is a unique opportunity for us. The use of contextual bandits in the targeting of our recommendations enables us to pursue completely new approaches in personalization by also personalizing the delivery of the respective recommender to the user. We have already achieved good results in our newsletter in initial experiments and are now working on extending the approach to the entire newsletter by including more contextual data about the bandits arms. Furthermore, as a next step, we intend to apply the system to our online store as well, in order to provide our users with an even more personalized experience. To build this scalable solution, we are using Google’s open source tools such as TFX and TF Agents, as well as Google Cloud Services such as Compute Engine, Cloud Machine Learning Engine, Kubernetes Engine and Cloud Dataflow.”—Christian Sager, Product Owner, Personalization ( Digitec Galaxus)

Since the existing architecture and system is also dynamic, it will automatically adapt to new behaviours, trends, and users. As a result, Digitec Galaxus plans to re-use the same components and extend the existing system to help them improve the personalization of their homepage and other current use cases they have within the company. Beyond clicks and user engagement, the system’s flexibility also allows for future optimization of other criteria. It’s a very exciting time and we can’t wait to see what they build next!

At 340 square miles, the City of Memphis is among the largest in the United States in terms of land area. Memphis has over 6,800 lane-miles of city streets, enough to drive back and forth to Los Angeles four times. Keeping these streets well maintained and safe for citizens and visitors is a major priority for the city.

Lots of traffic, lots of roads, and a four-season climate prone to wintertime freeze-thaw-refreeze cycles means the opportunity for potholes. Although the city aims to fill potholes within five business days of notification, it can take longer, especially during winter and early spring. Last year, the city’s Public Works crews repaired some 63,000 potholes, only 20% of which were reported by residents. Approximately 32,000-man-hours each year are spent repairing potholes, with seasonal fluctuations requiring ten to twelve Street Maintenance crews working steadily during the winter months. Still, many went unreported, leading the city to flag pothole request resolution under “needs improvement” on its open data portal website.

Like many large cities, Memphis also struggles with vacant and blighted properties. Nearly 15,000 properties in Memphis are likely vacant, and city officials contend that many are owned by out-of-town investors who live elsewhere and do not take necessary restoration or maintenance steps. These properties can decrease the value of surrounding real estate and discourage new businesses and other residents from moving to an area. Citizen frustration and concerns over the number of blighted properties has made blight eradication a major focus of the City of Memphis.

Historically, residents reported potholes and blighted properties by calling 311, or more recently by using the Memphis 311 app. However, these reports only covered about 20 percent of the problems — often the worst cases. And by the time residents took the initiative to submit a 311 report, they usually weren’t feeling good about the situation.

Recognizing that potholes and vacant properties are often the most visible indicators of whether a city government is doing its job efficiently, Memphis Mayor Jim Strickland and CIO Mike Rodriguez began looking for ways they could apply technology to fix the problems. Mike approached Google for ideas, and Google recommended conducting a machine learning proof-of-concept (POC) with SpringML, a Google Cloud Partner.

“Memphis is focused on easy living, and we want to do everything we can to keep our citizens happy,” says Mike Rodriguez. “Working with Google and SpringML to reduce potholes and urban blight using machine learning and artificial intelligence was an easy decision.”

Bringing machine learning to city operations and budgets

The city’s goal is to detect potholes and abandoned properties by analyzing video footage of roads and residential properties. It wanted to classify potholes by width and depth, and share the information with workers who can repair them. For abandoned properties, it wanted to enable more strategic deployment of resources for homeowners citywide and take action to hold neglectful property owners accountable.

The POC began by training TensorFlow models for ML object detection using preconfigured AI Platform Deep Learning VM Images on Compute Engine. SpringML helped set up cameras and developed a user interface to collect pothole data and automate the 311 ticketing process.

Together, the teams analyzed 30 days of video from a moving city bus and high-resolution video from 360-degree cameras mounted to a code enforcement vehicle, overlaid with data from 311 reports. As the models were refined, accuracy quickly climbed from 50 percent to over 90 percent as models were taught to differentiate a pothole from a manhole cover or other object.

The city also imported routes, potholes, and paving data along with geolocation data from ArcGIS and Google Maps into BigQuery to better understand street conditions and the proximity of potholes to one another. BigQuery also analyzes city property records, tax records, 311 reports, and third-party survey data on-demand to predict where homes are starting to become run down and where neighborhood decay is most likely to occur. The SpringML team created a pilot analysis to begin vacant property protections and developed a user interface tool to interact with the model’s results.

“Google Cloud Platform made it possible for us to experiment with machine learning and artificial intelligence to help solve our city’s problems while working within the budget constraints of a municipal IT organization,” says Mike. “Google turned a ‘nice to have’ into a ‘let’s do this!'”

Identifying 75 percent more potholes

Memphis expects to substantially reduce the number of potholes on its streets, creating a better driving experience for residents and visitors alike. Because drivers won’t be as likely to swerve to miss a pothole, streets will be safer and friendlier to bicycles and scooters. Fewer potholes will also save the city between $10,000 and $20,000 annually in city claims that it pays out in cases where vehicle damage results from a pothole that was not addressed in a timely manner.

“Historically, Public Works has relied primarily upon Street Maintenance crews to proactively locate and fill potholes. As Memphis has over 6,800 lane-miles of public streets, it is a daunting task to reliably survey the entire system in an efficient and systematic way,” says Robert Knecht, Public Works Director for the City of Memphis. “The outcome of the data collected will be invaluable to Public Works so that it can ensure it is managing the city’s street system in a more proactive manner.”

Memphis will be able to better prioritize road maintenance based on condition and impact, increasing the efficiency of its Public Works road crews. Analyzing video of streets also gave the city visibility into issues it wasn’t previously aware of, such as curbs, gutters, and manhole covers that had been mistakenly paved over and need to be excavated. The ML process is easily transferrable to other concerns as well, helping the city identify illegal signs or spools of cable hanging on light posts that could be potentially unsafe.

Helping communities recover and thrive

Memphis is also having success in analyzing predictive trends to combat high rates of abandoned and blighted properties, surpassing 97.5 percent accuracy. “In the past, Public Works experimented with comprehensive, city-wide blight identification by using approximately 200 volunteers to survey and photograph over 237,000 city parcels. This effort was costly, took a long time to complete, and resulted in inconsistent data collection,” says Robert. “Blighted property conditions can change quickly in a city the size of Memphis. Now, with this new technology, Memphis will be able to make a significant difference in the efforts to proactively and comprehensively identify and manage blighted and substandard properties.”

Code Enforcement with better data-driven detection mechanisms enables the city to also identify cases where homeowners are not physically or financially able to keep up with the challenges of homeownership and make them aware of resources that are available to assist them. Memphis Code Enforcement can do a better job of finding people living in derelict properties that pose hazards to inhabitants’ health and safety, and help them fix those problems or find a new place to live.

“Using SpringML and Google Cloud Platform to detect indicators of vacant or blighted properties will help Memphis create safer neighborhoods that will be more attractive to businesses and home buyers,” says Mike. “Property values and employment will go up, crime will go down, and social services can be more focused and effective.”

Revolutionizing service delivery for citizens

Memphis is proving the viability of a cost-effective, cloud-based machine learning model that other cities can follow. The city is already looking into new applications of AI and ML that will further improve city services and help it build a better future for its 652,000 residents.

As part of his commitment to a transparent government, Memphis Mayor Jim Strickland created an open data policy that commits to releasing raw data and sharing it with citizens in a variety of downloadable formats. Going forward, this transparency will help citizens understand how their needs are being served and uncover new, innovative use cases for AI and ML.

“Our goal is to become a smart city, and technologies such as Google Cloud Platform and SpringML put us ahead of the game,” says Mayor Strickland. “Google understands data, and there isn’t a better company to help us analyze our data resources for actionable insights.”