Brian Rojas

Did a talk at the IGDA TC Chapter about Building Excalibur.js! We dive deep into all the bits of making the game engine! https://lnkd.in/g5Vr_M9Q #excaliburjs #typescript #javascript #gamedev

6

I've been hard at work, getting ready to stand up Pixel Valley! Got a lot done on the conversation engine and created a quick StreamLit app to test it. Doesn't look like much, but I feel this is a big deal because it is "not just another chatbot". As you can see here, I am having a coherent conversation with an entire group of agents, that are each responding or not responding as appropriate. This was created with off-the-shelf gpt-3.5-turbo. Instead of using expensive fine tuning, it is accomplished using a combination of clever prompt engineering, RAG, and AI function calling. It's getting close!

11

One common question I get is, "What do you do as a Developer Advocate?" or "What does the Developer Experience team do?" (Even within Sendbird, haha 😂) The answer varies because the role of a Developer Advocate differs across companies. But I can clearly share what we do at Sendbird! And one great way to show this is by highlighting what we've accomplished!! Here's a recap of what we've done over the last three months. Huge thanks to the team! 🙌 (Jane Lee Katherine Na Yeon Kim Mansoo Kim Sunghyuk Kim Mickey C.) ======= 🚀 Dogfooding Session: We conducted an internal session for Sendbird Business Messaging, simulating a customer use case to improve our onboarding experience. 🎥 YouTube Videos: Released 5 videos covering AI Chatbot features to help users get started quickly. 📝 Feedback Channel: Actively collected and addressed feedback to enhance our documentation. 🤖 Interactive Demos: Built targeted demos for AI Chatbot in E-commerce, Healthcare, and Online Banking. 📚 Documentation Revamp: Improved our "Getting Started" guides, resulting in increased user engagement. For the full story, check out my Medium post: https://lnkd.in/gt-HGnER Looking forward to what we'll achieve next! 🙌 And one thing more, we're hiring a Technical Writer(https://lnkd.in/gFM4R9Wh), Please feel free to contact me if you are interested in this role! Also, if you're interested in becoming a Developer Advocate, don't hesitate to reach out!

53

2 Comments

Here's my first new talk in years, grown from one of the seeds Dragan Stepanović is always planting. It's for anyone who cares about software investments yielding returns. It's called "Not So #ExtremeProgramming": https://lnkd.in/ebywkddq #XP #XPLives #Agile #Scrum #LeSSIsMore

25

1 Comment

I’m intrigued by the internals of JAX’s transformations, such as jax.grad and jax.jit. JAX’s documentation briefly introduces the concept of a **Tracer**—a stand-in for JAX Arrays that records execution progress and captures information about intermediate results. Here’s a simple code snippet that reveals some intriguing properties of Tracers: 1. Each transformation relies on a specific type of tracer. For instance, jax.jit uses DynamicJaxprTrace, while jax.grad utilizes JVPTrace. 2. When we compose transformations, such as jax.jit(jax.grad(f)), the traces are composed—meaning the primals in JVPTrace are traced with DynamicJaxprTrace.

52

I'm expecting our new sync engine in @triggerdotdev to be done by the end of this week: 💨 Fan-out job per connection 🏦 Bank account job (balance and transactions) ⏳ Delay between accounts for rate limits 🔔 Webhooks for Plaid & Teller 🔄 Retries, error reporting, and logging

14

Tracer bullets and prototypes. Such a needed concept! I’ve had two company initiatives this quarter run over timeline. While there were a good amount of factors, I know two had part to play: 1. Making assumptions about things we don’t control (third party api’s in this case) 2. Not boiling down to the “minimum path to value” to get the feature out “early and often” The first could be solved with prototyping. The second solved with tracer bullet thinking. Prototyping and tracer bullets serve a similiar purpose in that they help you derisk projects by uncovering hidden assumptions as early as possible. The biggest difference is that prototyping gets thrown away and tracer bullet development does not. Prototyping might be architecture diagrams, curl requests to api’s, graphs, data visualizations or timing tests. It’s a sandbox method that lets you get dirty and get a handle on what you’re doing. But it’s never meant to get into production. It is meant to test your most basic and risky assumptions. Tracer bullets is the minimum line to production. How much can you pair down this feature to get the some amount of value out in front of your customers *as soon as possible*. What if I told you that instead of a month you have two weeks? One week? If you have a 5 layer archetcture, it’s a thin slice through all 5 layers. Note that you should still have all 5 layers *well defined*, they just don’t need to be all *filled in* yet. That difference is the key and takes a lot of skill to get right. Good tracer bullet execution means that when you go back to impliment the rest of the feature, it should feel like it slides right in. I’m working to grow myself in these things. I’m tired of the “feature being 95% done for weeks” syndrome. I’m hoping that getting better at these will ameliorate this issue.

5

Hey everyone! Wanted to share my first blog post about some of my work on Haskell Tooling at Mercury. We face a pretty unique set of challenges in terms of the scale within the Haskell ecosystem. Static-ls, which started as one of my side projects, has grown into one of the of the ways we handle that and become a tool used by numerous backend engineers at Mercury. Please give it a read if you're interested!

88

11 Comments

ruby-lsp is a solid extension to use with vscode. I spent some time today with it debugging a couple of complex issues and I was able to dig deep into our dependency tree easily without having to manually search for things. Definitely give it a shot if you're using #ruby #rubyonrails.

1

As I'm getting back into #Angular, I'm remembering how much my brain has trouble working with RxJS Streams. But, it just occurred to me, I don't actually have to use their HttpClient module—I can use my own fetch client. All I want is Promises and AbortControllers. This is not a slight against RxJS—I know that many people are intensely in love with it. But my brain just can't deal. Resolves and rejections for HTTP requests, that's what my brain can understand. RxJS feels like more abstraction than I can handle. Maybe it has something to do with my Aphantasia.

5

6 Comments

Working on a specification for https://lnkd.in/g-pBfAQv. Then I can verify it with tests and https://blprnt.tech/. I'm reminded that wormwood is optimized for readable source code, not UX. There is very little syntax validation.

4

I wrote a thing about zero-cost abstractions, with a real usage example: https://lnkd.in/g4mSJ52H This is programmer specific, and fairly "beginner level", so feel free to ignore if you're not in a target demo where that might be interesting to you. I wanted to write the blog post, based on some recent interactions, and figured I'd link it here in case it's interesting to anyone it might be shown to.

3

Gotta say, I agree with a lot he had to say. Now, I’m not going anywhere. Been in mobile dev since 2008 and will be staying. But I see more and more as I get exposed to newer projects that I’m no longer cut out for this. Not that I’m not cut out for mobile dev, I’m not cut out for the over-engineered, overly complex projects that proliferate nowadays. I now see that my space is developing new projects using the same code, methodologies and design patterns that I used 4, 5, 6 years ago. #iOSDev

13

Recently, I've become the lead dev for our team. That's mostly meant a lot of meetings, and second mostly meant a lot of whack-a-blocker, but third mostly it has meant confronting an entirely new kind of imposter syndrome. I often think of the words I heard Ty Carlson utter once when he was asked if he wanted to join a random meeting, "Sure, I like... knowing things." Me too - I like thinking through contingencies and "being looped in" and diving deep and other general anxiety-reducing activities. And so, trying to do that across several large multi-sprint projects of varying (high-to-extreme) complexity and time-sensitivity all with cross-team functionality and moving parts changing by the minute - well, that led to a certain self-imposed pressure (and ensuing face-planting failure) to always know the right answer, thinking that was the best way to have our team's developers' backs. This felt especially true since I was following Andrew Kirchmyer - who does, in fact, always know the right answer. I'm finding a lot of freedom lately in realizing that it isn't about having all the answers - it's about collectively creating an environment where the right answers can emerge (the right people, the right scope, the right next steps, etc.) and then defending that against all the various pressures and subtle insecurities and inconsequential sink holes that can challenge it. I've probably flopped on that with my first few projects, but I'm looking forward to growing in this role with that focus. It makes things more fun, and, hopefully, will let each of our developers flourish and grow as well.

67

17 Comments

There’s something for everyone in web development in here:

8

1 Comment

🎺 Feature Friday (is that a thing? I'm running with it...) New on capture.dev this week - take your product feedback beyond random screenshots and use your camera and mic to add extra context to those more difficult to explain issues. Think of it like a Loom, except: 👉 Devs get all of the technical info they need, like network requests and console logs, for zero extra effort. 👉 Record directly in your browser, with no extra software or extensions needed Free to try now over on capture.dev

14

A lot has happened since I last posted about my hobby project Trickster's Table over a year ago. The app has gotten some good press and very good reviews on Google Play and the Apple App Store. * Dice Tower review by Chris Yi: https://lnkd.in/gVNdwYuh * 4.9/5.0 rating on the Apple App Store: https://lnkd.in/gmKhFVGw * 5.0/5.0 rating on Google Play: https://lnkd.in/gnCSfn3x Since my last post here I've added 5 new licensed games: • Yokai Septet 4-player by Muneyuki Yokouchi - a very strategic partnership game where you are trying to capture sevens • Splitter by Stefan Nikolić - a roll and write game where you score by grouping together numbers (two 2s, three 3s, etc.) • Short Zoot Suit by Taylor Reiner - a trick taking game where you try to balance short suiting and winning tricks • Dealer's Dilemma by Shreesh Bhat - a streamlined precision bidding game where you bid with the cards in your hand • Hotdog - a 2-player bidding game where players can bid to select if high or low cards win, trump, and a special rank I started experimenting writing the games in Rust earlier this year and the experiment paid off. The compiled Rust code can run many more simulations than Dart (which I had used before to implement the games) in the same time. This reduces the amount of time it takes to get high level play implemented compared to training a neural network to play the game. The level of play has improved a lot in the newer games and I'm going back and rewriting the old game engines in Rust between adding new games. The attached picture is a search tree for Hotdog where the bot is deciding how to bid. You can find the source code for the newer games written in Rust here: https://lnkd.in/g5iQ9tyw If you’d like to support or just follow along, check out https://lnkd.in/g6Akd3zS. Members can vote in the new game polls to help shape what’s next for Trickster's Table! 50% of donations are shared with designers/rights holders of games that appear in the app (divided by play time) and 50% of donations are used to fund the development of new games.

21

1 Comment