Sentitrac Beta Launch

You are reading "Software, My Dream Job," a software engineering newsletter covering one person's perspective from working in the software industry. If you'd like to receive notifications for new editions, click or tap "Subscribe" above.

First off, hats off to the Miami Heat, who, in winning Game 2 of the NBA finals, bought me an extra couple of days to release Sentitrac before the finals are over!?

With Sentitrac's release around the corner, I wanted to share how I built it over the past couple of months.?

Focus

A big part of my approach, which was heavily time constrained since I work full-time at LinkedIn, was to focus, focus, focus.

That focus started with using what I'm familiar with. I've been playing with Django for more than 10 years and have had apps on Heroku for at least that long too.?

Backend framework and programming language: ??

Hosting stack: ?

A quick aside here: I started building directly on AWS, and within just a handful of days realized the mistake that would be. While more expensive, Heroku provides so much "for free" that it lets me focus on what's important to me and what I'm building. I'll happily pay to avoid a multi-week rabbit hole into AWS configurations.

Another big area of focus was not building my own authentication or billing system. Auth0 and Stripe, respectively, provide best-in-class services for this exact purpose and sure, it's not free, but my time isn't free either!

Not only does this save me time, but it gives me peace of mind. By paying these service providers for these critical components, I can rely on their expertise to help me instead of having all that stress on my shoulders by building it myself.

Dash of New

As much as I wanted to avoid newer frameworks, Next.js stood out as an improvement over what I'd used before with React. With so much material online that would help me, I knew Next.js could work. It would also be fun to learn something new!

Of course, there's also the new type of data I'm bringing to market -- sentiment data for sports! Over the past few weeks, I've started gathering daily sentiment for all NBA teams and have been thoroughly impressed by the ability of OpenAI's models to extract sentiment and provide summaries. There have been some interesting hallucinations; for example, the Nuggets got a poor sentiment score after losing Game 2, so much so the model thought they were down 0-2.?

Speaking of the model, I'm currently using GPT-3.5-turbo as the primary language model, providing both sentiment and summarization. GPT-4 is easy to switch to, but with a cost ~50x greater than GPT-3.5-turbo, it's not yet sustainable for me to use that as the primary model producing sentiment for Sentitrac.

While I'm "only" processing the ~120 teams across the 4 major US sports leagues, player data is next! Combined with pulling in data from Twitter, player sentiment will be a crucial feature for Sentitrac for the full product launch later this summer.?

Why is it not part of the beta, then? Well, there are many more players than there are teams, and while I've used Twitter data before in my thesis, it's one of the more complex data sources to work with. That said, I wanted the beta to prove some of the basics and get the overall application working (IE account setup, subscriptions, referrals) first while I iron out the kinks of player data a bit further.?

?

Sentitrac

The beta launch is scheduled for next week, at which point the most basic version of Sentitrac will be available to subscribe to. You'll see sentiment data on all major US sports teams (NFL, NBA, NHL, and MLB). That means a current sentiment score (on a scale of -10 to 10), a histogram showing sentiment over the past few weeks, and an AI-generated summary of the team's current sentiment.?

Player data and sentiment will launch in July, but subscribing to the beta will be your best option if you want to secure the most significant discount on Sentitrac. The beta will be a flat $0.99/month, and I expect to charge between $35 and $100 per month once Sentitrac gets out of beta.?

Of course, subscribing at launch will also get you a discount. I'm a big fan of rewarding early adopters, but the longer you wait, the lower your discount. The only exception will be for those who are great referrers -- the details of the subscription referral program I'm still working out, but in spirit, if you're referring more than 10 other subscribers in a year, you'll get that year for free. Start thinking of who you'll want to invite!

As with any beta, please reach out if there are features you'd like to see or sports, teams, or players you'd want to track. Any and all feedback is greatly appreciated, and I hope you find value in what I've built!