Joshua Apstein
Northridge, California, United States
3K followers
500+ connections
About
Software Engineer looking for new opportunities. I love solving problems. Getting stuck…
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
-
Licenses & Certifications
-
CAMTC Massage Therapy Certification
CAMTC
Credential ID 62377
Projects
-
Vicissitutor
- Present
See projectSocket Based App in which an instructor can control game settings of a student
- Moved image rendering to server using web sockets and Phantom.js to decrease load on client device
- Built dynamic controller page to make for easy development and loading of new games for the platform
- Set up web socket ledger limiting web socket connections to two unique users each with distinct roles
- Integrated mobile browser API to incorporate user input from mobile devices into desktop game…Socket Based App in which an instructor can control game settings of a student
- Moved image rendering to server using web sockets and Phantom.js to decrease load on client device
- Built dynamic controller page to make for easy development and loading of new games for the platform
- Set up web socket ledger limiting web socket connections to two unique users each with distinct roles
- Integrated mobile browser API to incorporate user input from mobile devices into desktop game application
- Structured test suite to test multiple clients using ZombieJS to keep code clean, testable, and dynamic
- Deployed NoSQL database on EC2 and application using Elastic Beanstalk requiring configuration for web sockets
-
Gooey Pages
-
Generates source code for full-stack project with one click
- Led small team in building out a new front-end to allow users to drag and drop HTML elements into a template
- Built core implementation of the algorithm used to turn this HTML template into a full HTML file
- Refactored code from jQueryUI to gridsterJS for more friendly UX and smoother Drag and Drop functionality
-
Ping-Pong Ladder
-
See projectRanking system for Codesmith Staff and Students’ Ping-Pong skills
- Developed underlying algorithm in keeping ratings based on chess ELO ratings
- Used promises and Node’s event loop to manage simultaneous asynchronous updates from multiple users
- Improved site load time by 84% by condensing and caching images (benchmarked with GTmetrix)
- Deployed on Heroku with MongoDB for Codesmith staff and students to use immediately
-
Pleated Slacks
-
Configuration tool for Docker
- Built algorithm to build Dockerfiles based on modules in your package.json, and matched with official Docker images
- Conducted unit and integration testing to ensure all of our code was working before new changes - could be added
- Established command line tool for simple User Experience and ability for quick deployment
Languages
-
English
-
Other similar profiles
Explore more posts
-
Mohamed Mostafa
Software engineers fear the AI that will take their job while the real danger I see should be in companies and solution owners themselves. Products that you spent years and tons of dollars for buildng selling it for whatever the number of subscriptions is can now be easily cloned and it costs nothing but the Gen AI subscription and some small resources. I am really curious of how the scene will look like after maybe Q2 of 2026. The development that business owners needs to follow “not to fail in the common nokia example” is now talking days not years like all used before!! #ArtificialIntelligence #GenerativeAI #SaaS #StartupLife #ProductManagement #SoftwareEngineering #TechTrends #DigitalTransformation #Innovation #FutureOfWork #Entrepreneurship
2 Comments -
Valeria Rodríguez
I always tell devs: if you're stuck on what to build with AI, reverse-engineer real tools. Try this: - Look at GitHub Copilot—what does it automate? - Use Grammarly—how does it rewrite? - Play with Notion AI—what prompts power it? Then, recreate a tiny version of one. That’s the best way to learn.
-
Anvesh Mora
𝗕𝗶𝗴 𝗗𝗿𝗼𝗽𝘀 𝗳𝗿𝗼𝗺 Vercel 𝗮𝘁 𝗦𝗵𝗶𝗽 𝟮𝟬𝟮𝟱 – 𝗪𝗵𝗮𝘁 𝗬𝗼𝘂 𝗦𝗵𝗼𝘂𝗹𝗱𝗻’𝘁 𝗠𝗶𝘀𝘀 Vercel just announced some major updates at Ship 2025. whether you’re builder with the “build everything on Vercel ” team or still nursing a love-hate relationship, these two announcements are definitely worth your time: 𝟭. 𝗦𝗺𝗮𝗿𝘁𝗲𝗿 𝗖𝗼𝗺𝗽𝘂𝘁𝗲: 𝗙𝗹𝘂𝗶𝗱 & 𝗔𝗰𝘁𝗶𝘃𝗲 𝗖𝗣𝗨 𝗣𝗿𝗶𝗰𝗶𝗻𝗴 This is a big deal. Vercel now charges based on actual CPU time—only when your code is running. No more paying for idle time. It’s similar to how Cloudflare Workers charge, and makes the pricing model more transparent and developer-friendly. Cloudflare was the only major player doing type of service and price model, and It's good to have one more major player in the space. I think this would stealer idea to implement in Kubernetes echo-system. Read more: https://lnkd.in/gpUcCMb9 𝟮. 𝗩𝗲𝗿𝗰𝗲𝗹 𝗦𝗮𝗻𝗱𝗯𝗼𝘅: 𝗥𝘂𝗻 𝗨𝗻𝘁𝗿𝘂𝘀𝘁𝗲𝗱 𝗖𝗼𝗱𝗲 𝗦𝗮𝗳𝗲𝗹𝘆 This not a new concept if you know how platforms like LeetCode or Replit does the code execution and running tests. Vercel introduced a secure sandbox for running untrusted or AI-generated code to make sure "Code generated by an AI agent doesn't compromise your Infrastructure". More details: https://lnkd.in/gP72t8Zp I personally loved these features. 𝗩𝗲𝗿𝗰𝗲𝗹 𝗶𝘀 𝗻𝗼 𝗹𝗼𝗻𝗴𝗲𝗿 𝗷𝘂𝘀𝘁 𝗮 𝗳𝗿𝗼𝗻𝘁𝗲𝗻𝗱 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺. These drops show they’re building for the next wave of apps—AI-native, serverless, and secure by default. Share your thoughts in the comments section. Are you excited or what?
-
Pranav Patel
🔥 You burn $1M worth of Sonnet 4.5 tokens monthly, when simple ReACT agents can get you same performance for like $100. If you don't know what ReAct is, and burn "thinking" tokens cuz you think that's the only way to make the model smarter. You are stupid. Read on. You need this. ReAct prompting technique combines the “reasoning” and “acting” capabilities of an LLM to help with tasks like action planning, verbal reasoning, decision-making. It does so by forcing the model to reason and observe before acting. This helps the model to analyze the context and situation so far, and then take the actions necessary to move forward. Now current agents behave similarly, but they use so much more tokens to "reason". It breaks down the task, analyze, and then proceeds with tool calling and response generation. But most of the time this reasoning is simply not needed. You can tell the model "how to think" and "what to think about". This constrains the model to only focus on the most importing things. The goal here is, to do structured thinking, rather than open loop thinking. ReAct prompting helps with this and gives a very good framework on how to integrate tool use and interleaved thinking and user response generation all in one go. The best part is that you don't even need a huge reasoning model for this. Basic LLaMA models can be prompted to do this and would give a massive boost in performance. PS: Chain of Thought is a very strong contender, but its tricky to implement along with tool calling and interleaved thinking specifically. - ReAct provides a more structured option out of the gate. Let me know if yall need any help with ReAct prompting and saving on token costs with your agentic workloads. Peace.
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content