React Conference Talks and Workshops

Upgraded a ReAct agent with 𝟳 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗽𝗮𝘁𝘁𝗲𝗿𝗻𝘀 + 𝗮 𝘀𝗲𝗹𝗳-𝗶𝗺𝗽𝗿𝗼𝘃𝗶𝗻𝗴 𝗹𝗼𝗼𝗽 — each designed to fix a real failure mode. 𝗠𝗲𝗺𝗼𝗿𝘆, 𝗿𝘂𝗹𝗲𝘀, 𝗮𝗻𝗱 𝗽𝗿𝗼𝗺𝗽𝘁𝘀 𝗲𝘃𝗼𝗹𝘃𝗲 𝘄𝗶𝘁𝗵 𝗲𝘃𝗲𝗿𝘆 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻. Not built on frameworks. No DSPy. No LangChain. Just raw JavaScript — so every failure and every fix is visible. Here’s what actually made a difference: • 𝗖𝗵𝗮𝗶𝗻 𝗼𝗳 𝗧𝗵𝗼𝘂𝗴𝗵𝘁 — forces structured reasoning • 𝗦𝗲𝗹𝗳-𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 — reduces randomness, improves reliability • 𝗣𝗹𝗮𝗻 & 𝗘𝘅𝗲𝗰𝘂𝘁𝗲 — breaks down complex tasks • 𝗧𝗿𝗲𝗲 𝗼𝗳 𝗧𝗵𝗼𝘂𝗴𝗵𝘁𝘀 — explores multiple paths, not just one • 𝗦𝗲𝗹𝗳-𝗥𝗲𝗳𝗶𝗻𝗲 — critiques and improves its own output • 𝗥𝗲𝗳𝗹𝗲𝘅𝗶𝗼𝗻 — learns from mistakes • 𝗠𝗲𝗺𝗼𝗿𝘆 — adapts to user context over time • 𝗜𝗻𝘁𝗲𝗻𝘁 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 — improves how the system understands the ask The real shift? → Moving from “generate an answer” → To “design how the system thinks, learns, and improves” That’s where agents start becoming systems, not scripts. #AIAgents #GenAI #AIEngineering #LLM #JavaScript #ProductThinking #SelfLearning #AI Git: https://lnkd.in/gmKaptts

🚀 Excited to share my latest project: RouteNav — an AI-Driven Traffic Optimization System! I built this full-stack application to solve real-world routing lag and deliver fast, efficient navigation. How it works: Backend: Built with Node.js, Express, and SQLite, featuring secure JWT authentication. Speed: Uses React and Leaflet.js with a smart strategy to load driving paths instantly (1–2 seconds) without lagging the UI. Live Tracking: Streams live GPS data via the Geolocation API. By optimizing updates, it keeps the computer CPU usage under 10%. Smart Features: Cleans up GPS signal jitter using a rolling buffer and includes Web Speech API voice control. 🔗 Check out the source code here: https://lnkd.in/gzApKkNW #FullStack #ReactJS #NodeJS #WebDevelopment #CodingProject

We've added new categories to the Code Arena: Frontend leaderboard, covering 7 domains of agentic web development. In this clip, Aryan Vichare and I-Hung Hsu discuss how the new categories surface a more detailed picture of which model fits which use case. The full video on YouTube covers the ML methodology behind the rankings, what the shifting data shows about how people are building with AI for the web, and which models are leading in specific niches: https://lnkd.in/gS5ZigNs

Your AI agent ships React code quickly but often it's bloated, slow, and contains hidden mistakes. React Doctor v2 addresses these issues before they reach your users. It is compatible with Next.js, Vite, and React Native, and can identify performance problems, bad patterns, unnecessary re-renders, and broken architecture in just minutes. Simply run: npx react-doctor@latest. This tool is open source and extremely beneficial for anyone working with React. 🥳

I recently ran a Next.js pull request through my reviewer, which identified two significant issues: - Using useState inside a Server Component - Using getServerSideProps inside the App Router While both issues passed code review, they would cause failures in production. What surprised me most wasn’t the bugs themselves, but how easy they are to overlook. Typically, PR reviews emphasize: - Syntax - Logic - Readability However, many real Next.js bugs stem from boundary issues: - Server vs. client - App router vs. pages router These distinctions are not always clear in diffs, particularly in larger pull requests. This is the challenge I aim to address with MergeWell. It’s not just about generic AI code review; it’s about catching these framework-level mistakes before they reach production. I also experimented with a lighter, open-source version, which can be found here: https://lnkd.in/g4MhF4yt. I’m curious to hear from others—what’s a Next.js bug that slipped past your code reviews and only appeared later? For more information on MergeWell, visit: https://lnkd.in/gzGVBqXv

I stopped arguing about Claude Code vs Codex. Now I use both. Codex feels better when the change could break stuff & involves multiple repositories (gpt-5.5) Claude Code feels better when the work needs taste like new UI or web design (opus-4.7) So my workflow has become: 1. Codex maps the system 2. Codex handles the architecture + backend changes 3. Claude Code cleans up the UI 4. I review the diff before shipping Most people are trying to pick one AI coding tool. I think it is knowing which tool to hand the problem to.

Work on source card has halted, in the age of AI you have even less of a reason to ship slode(sloppy code) and for that reason im going deeper into the javascript to understand why my app does what it does, much of the HTML and CSS comes from me but like 90% of the JS is AI generated for source card, but not to worry v3 will come out with even more features and v4 and v5 will ship even faster now that im more confident in my code, I just want to make sure people have the best card creator out there

This is an absolute utility of a product conceptualized on the core concepts of modern AI... I will definitely recommend this for organisation and individuals to explore this

Swaya.me is now open source. What started as an idea grew into a full-fledged real-time quiz and assessment platform with AI-powered analysis, live engagement, proctoring, multilingual support, and a scalable multi-tenant architecture. Built with: • FastAPI + async SQLAlchemy • React 18 + Vite + Ant Design • Real-time WebSockets • AI-driven response analysis • Exam-mode proctoring • Support for 11 languages Now the entire project is available for the community to explore, learn from, contribute to, and build upon. Star the repo, fork it, or contribute: https://lnkd.in/gV3pqcDU

🚀 Day 2 of Learning & Implementing RAG: Bringing the Data to Life! Yesterday, I built the base data ingestion pipeline. Today was all about bridging the gap between retrieved vectors and real-time generation. Day 2 is officially complete, and our RAG system is fully alive! 🧠✨ Here is the progress report: 1️⃣ The Generator (Groq + Llama 3.1): Integrated the Groq API running Llama 3.1-8B-Instant. By passing our retrieved context chunks from Qdrant directly into the LLM prompt, we're getting hyper-accurate, context-grounded answers. 2️⃣ Server-Sent Events (SSE) Streaming: Static, slow REST responses are out. I built SSE streaming on the backend! It works in two phases:   👉 Phase A: Instantly stream the context citation references to the client so the user sees source documents immediately.   👉 Phase B: Stream the synthesized answer token-by-token for that ultra-smooth, real-time typing interface. 3️⃣ Production-Grade Validation: Added Zod schema validation to secure and type-check the request payloads. 4️⃣ Client-Side Security (Flutter): Integrated secure device storage (`secure_storage.dart`) to store API credentials safely on the app. 💡 What I learned today: Streaming responses isn't just a UI detail—it completely changes the user experience by reducing perceived latency. Designing SSE connections between a TypeScript backend and a Flutter client takes some parsing care, but the real-time UX is incredibly satisfying! Day 3: Parsing the live stream on the frontend and building an interactive dashboard. How do you handle real-time streaming connections in your AI apps? Let's discuss in the comments! 👇 #RAG #ArtificialIntelligence #TypeScript #NodeJS #Flutter #SSE #Streaming #LLM #Groq #Llama3 #BuildInPublic #SoftwareEngineering