Fullstack Engineer

At VoiceOps (YC W17), we’re revolutionizing the $500bn+ call center industry by giving managers AI-powered tools to level up their sales reps. We're growing fast - we just raised a $9M series A.

CORE TECHNICAL CHALLENGE

Human conversations are messy. Previous software attempts at structuring conversation data leave a lot to be desired.

We are taking the same approach to conversation analysis that Uber/Cruise/Waymo are taking for self-driving cars. Building troves of training data, and solve lots of edge cases piece by piece towards the goal of having an incredibly reliable system.

Our core technical challenge is how to take billions of audio recordings of messy, unstructured human conversations and make sense out of that data in a way that is: a) accurate b) cost efficient, and c) highly scalable.

(The corresponding product problem is how to take well-structured data and make it actionable for the end-user)

TEAM AND PROCESSES

We are currently a 5-person team that includes engineering, data science, design, and sales. We start the week with a product planning meeting where everyone gets together with our cofounder Nate to set the product plan. We optimize for tight execution on a limited set of priorities, which creates a collaborative environment (we're often all working together on the next big user-facing project together) and mutual accountability.

WHY JOIN US

Impact: the team is small, you’ll help set the template for engineering quality and process. You'll also be encouraged to contribute to product decisions, and help shape the direction of the company.

Stellar Team and Culture: We have very high standards for our engineering team, so you'll get to work with some of the smartest people you've ever worked with.

Growth: We've just raised our Series A and are looking for more smart people to continue growing quickly. You’ll be one of the first few employees and have opportunities to be a leader on a growing team.

PROJECTS YOU COULD WORK ON

1. Build a model to predict where a sales behavior should have happened We currently show managers which transcripts missed a key behavior, such as a close attempt. But managers then have to look through the transcript and decide for themselves where it should have gone. We want to save them the effort by showing them exactly where in the transcript we think the close attempt should have happened, so that all they have to do is leave a note for the Ashley saying “Ashley, here’s exactly where you could have gone for the close.” This will be a magical experience for managers and takes advantage of all of our incredible labeled data.

2. Train custom language models using our massive set of pristine transcripts Towards our goal of increasing margins (and accommodating a 6X increase in data processed in 2019), one lever we can pull is to make the automated transcripts more accurate, such that our team in the Philippines edits fewer snippets per call. Build custom language models for each customer. We have the benefit of lots of pristine, human-edited transcripts for each customer already, making this a promising strategy.

3. Run A|B tests to improve workflow efficiency for 1,400 transcribers The other lever we can pull towards increasing margins and accommodating 6X scale is to improve the workflow of our transcribers. We’ve built, in-house, software for them to edit transcripts, as well as an A|B testing framework to test workflow improvements. We need to extend this framework and do more experiments in order to make the transcription that workflow easy and quick. By doing so, we can help our transcribers edit each audio minute in less time, increasing our efficiency.

4. Migrate our codebase over to TypeScript We’d like to move parts of our codebase into TypeScript so that we can take advantage of static type-checking. Setting this up would be a great way to get your hands dirty with our codebase right away and be an important long term investment. We’re technology agnostic, so if you think there are better options here, we’re open to hearing your case.

5. Build out our microservices We’re dominantly a Rails and React shop, but we use Go for our heavier routes. We’d like using VoiceOps to be a smooth, fast experience for our customers and think we can push the needle forward here. You’d be able to come in and start impacting the speed at which we’re serving data to our customers on week one.