AI in tech: automation through machine learning
Automation is the most valuable time-saving device. Every automated manual process is time made available to pursue creative and product-defining ideas. Automating the simple stuff is easy. You’ve probably taken care of that already. But what if the task you’re trying to automate is not only slow but inaccurate, even when outsourced to a team of people? These are the kinds of problems that excite us at super.AI. We specialize in breaking down complexity and making difficult tasks seem simple.
We’ve helped tech companies to improve their throughput, hit tight project deadlines, and smoothly scale their products through the use of data programming. In this post, we show you how data programming can turn your seemingly impossible headache into just another background process.
Data programming
Data programming is our solution to the complexity of the real-world problems faced by people in the tech industry. You can think of it like the assembly lines that Henry Ford used to revolutionize the auto industry. It’s predicated on the idea that breaking enormously complex tasks down into their constituent parts diminishes the skill level and the manual labour required on the part of the individual.
The process we’ve developed means that problems that could previously only be completed by an expert or a large, expensive ML model managed by a PhD-educated data scientist are now simplified down to a task that anyone with basic training can handle through an intuitive UI.
At super.AI, we’ve worked with customers from across the tech industry to solve problems with unique complexities, including creating and maintaining dense product taxonomies and matching user queries with their intent in order to surface the most relevant support content. We’re going to walk through one use case to illuminate the way in which super.AI simplifies complexity.
LogMeIn: scaling for product launch
LogMeIn is a $1bn provider of SaaS and cloud-based remote connectivity services for collaboration, IT management, and customer engagement. They were looking to launch a meeting bot that could summarize and share meeting notes. This required automated understanding and summarization of natural language.
Growth pains: not scaling fast enough
At the outset, they spent most of their time labeling data from their customers’ meetings themselves. But it became clear early on that they had a choice: they could take technical shortcuts and compromise user experience in order to save time, or they needed to find a new approach:
We were starting to get frustrated just labeling meeting data with our own team. We realized we needed to scale our ML and human computation efforts.
Scaling with Super.AI
The solution we arrived at for LogMeIn was a string of data programs working in tandem, one taking the output of the other as its input:
- Audio chunking: first we broke up the audio recordings in smaller chunks of around 5 seconds
- Whitelisting: we used these short audio files as the input for our audio categorization data program, which sorted the clips into either the “whitelisted” or “not whitelisted” categories
- Transcription: we used the whitelisted audio files as the input for our audio transcription data program, adding context to the meaningful information (who said it, what is it about, when did they say it, why?)
- Summarization: given the annotated transcription, we produced a summary of the clip using our text structured data program
- Chunk amalgamation: we took the resulting collection of summaries and combined them into the final meeting summary to send back to LogMeIn
Results
Super.AI provided a solution that LogMeIn could scale up and down as their throughput demanded. Our production API allowed them to directly connect their product to our service and receive the results with guaranteed latency. They used our solution to scale their production traffic by 1400 percent. All of this meant that the team freed up valuable time to develop new product features and prepare for the solution rollout to its customers.
Software startup: data programming is adaptable
We made use of three of our data programs to solve LogMeIn’s problem, which covered both audio and text inputs and outputs. But this is only a small fragment of the full range of possibilities that data programming provides. A data program can take any input, whether audio, text, images, or videos.
We’ve helped a stealth software startup automate web scraping to build their software database on things like product pricing, feature existence, site design, and integration capacity. This web-scraping process had proved costly and slow, but super.AI helped not only to automate it, but to provide a 10-fold increase in throughput, from 30 to 300 products daily. Accuracy of the automated output hit 90 percent, compared to the previous levels of 30–50 percent. Additionally, we were able to provide information to the customer that was not captured before. The structured output we provided enabled improvements in their statistics and faster iteration.
No matter what your project requirements, data programming is flexible enough to accommodate them. If you want to set up AI or machine learning projects at your company talk to one of our experts, who can advise you on the optimal way to create your own assembly line.