Zach X

𝗧𝗵𝗲 𝗯𝗲𝘀𝘁 𝗔𝗜 𝗰𝗼𝗱𝗶𝗻𝗴 𝗹𝗲𝘀𝘀𝗼𝗻 𝗰𝗼𝘀𝘁 𝗼𝘂𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿 $𝟲𝟬𝟬 𝗮𝗻𝗱 𝗮 𝗺𝗮𝗿𝗿𝗶𝗮𝗴𝗲 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁. Our VP of Engineering, Xiaofan(James) Luan, was supposed to buy his wife a Dior bag for their anniversary.  Instead, he bought three Claude Code subscriptions and spent the holiday trying to cross-compile 2 million lines of C++. Every fix on one platform broke two others.  $600 later, the only output was "git reset --hard" — and a very cold dinner table.😂 "Make it compile on Windows" is a trap. The real goal was "compile everywhere without hacks" — no AI is going to figure that out for you at 2 am. What worked: constraints before code, review tests not code, bottom-up, one layer at a time. Same task, two days. Then he ran six parallel Claude sessions across three machines with git worktree. The bottleneck stopped being intelligence and started being how fast one person can alt-tab. AI solves exactly the problem you give it. Engineering is in knowing which one to give. His wife is still waiting for that bag. Full story: https://lnkd.in/gtsW_Wvk ——— Follow Milvus, created by Zilliz, for everything related to unstructured data

9

🚀 Excited to announce the launch of Shamsi Calendar API! 🎉 After weeks of development, I'm proud to release a comprehensive RESTful API for the Afghan Shamsi (Solar Hijri) calendar - now live at https://lnkd.in/g7TNWXys Why I built this: As Afghan developers, we've long struggled with the lack of reliable tools for our calendar system. This API changes that. 🌟 Key Features: ✅ Accurate Shamsi ↔ Gregorian date conversion ✅ Full calendar views (day, week, month, year) ✅ Complete holiday management system ✅ Trilingual support: Dari, Pashto & English ✅ Interactive Swagger documentation ✅ 151 tests with 85.6% code coverage ✅ Free and open-source 🛠️ Built with: TypeScript, Express.js, tested with Jest, and deployed with enterprise-grade security (rate limiting, CORS, Helmet) 📊 What you can do: Build Afghan calendar apps without reinventing the wheel Integrate Shamsi dates into your existing applications Access accurate holiday information Use it in any language: TypeScript, Python, Java, etc. 🎯 Perfect for: Afghan tech startups, government systems, educational platforms, mobile apps, and anyone serving the Afghan community worldwide. Try it out at the interactive documentation and let me know what you think! Contributions and feedback are welcome. #OpenSource #API #AfghanTech #ShamsiCalendar #WebDevelopment #TypeScript #SoftwareEngineering #Afghanistan

22

1 Comment

Hardware engineers right now: business as usual. Software engineers right now: refreshing LinkedIn every 30 minutes. AI can generate code. AI can't manufacture chips. The irony: the people building AI infrastructure have job security. The people using AI tools to be more productive are getting cut. Physical constraints protect jobs. Digital abundance doesn't. The market is telling you where to invest your skills. Visibility matters. Relationships matter. Having options before you need them matters most. Source: https://lnkd.in/gahzATXv

3

If you're a Senior Engineer looking to move up, the next role will likely be a Staff Engineer. And in this guide, Shruti shares tips from her own experience of getting promoted to Staff Engineer at PayPal and Slack. She talks about what Staff Engineers do (and how it's different from Seniors), why you might not be getting promoted, and how to take that next step. https://lnkd.in/g58dnFEG

103

3 Comments

Want to truly understand how AI agents work? Build them yourself. I found a repository that teaches you exactly that—without LangChain or any abstractions. It focuses on raw code and real understanding. The course consists of 9 chapters that build on each other: - Loading local LLMs - Prompt engineering - Function calling (tools) - Persistent memory - ReAct reasoning loops Here's why this matters: Frameworks are beneficial until something breaks. When that happens, you're left debugging code you never fully understood. This repository addresses that issue. Start by building everything from scratch. Then, when you later use LangChain or CrewAI, you'll have a clear understanding of what's happening underneath. Check it out here: https://lnkd.in/gy_hVF6r

20

Many are heralding large language models (LLMs) as transformative forces in our world. While I acknowledge the significant advancements they represent, I believe LLMs are not the ultimate goal for artificial general intelligence (AGI). For years, we have relied on neural networks, which serve as the foundation for LLMs. We have developed architectures that predict outcomes based on data to support decision-making. Additionally, we have refined datasets to provide context for various problems. However, the ability of LLMs to generate text and code does not equate to true intelligence. My skepticism stems not from a lack of potential in these models, but from the understanding that they are fundamentally similar to the algorithms that preceded them. AGI will not and does not stem from the LLMs we see today. LLMs are simply machines that have been trained on large datasets to determine the best word to place in front of the string of words they have already generated. Please do not be mistaken in thinking your LLM is somehow intelligent beyond making some significant mistakes.

9

🦺 🛟 Agent Safety > Model Safety LLMs didn't break your app. Agents can. As we wire LLMs to tools, browsers, payments & code the risk shifts from bad text to bad actions. Top failure modes (seen in the wild & in papers): 🚩 Prompt/Content Injection -> Tool Misuse. Malicious web pages or data instruct agents to steal secrets or make unintended API calls. This is now a primary risk for browsing and MCP - style tool agents. 🚩 Over permissioned tools & long-lived tokens. Agents get "god mode" scopes. A single injection becomes account takeover or irreversible ops. Security notes for MCP emphasise least privilege and short-lived auth. 🚩 Unsafe web autonomy. Benchmarks show web agents will attempt harmful tasks unless constrained (posting misinformation, illicit sales, etc.). You must measure misuse not just assume guardrails. 🚩 Supply-chain & retrieval poisoning. Agents trust plugins, third-party tools & indexed data that an attacker can taint. OWASP’s newest GenAI Top 10 calls this out explicitly. 🚩 Process safety gaps. NIST's GenAI Profile (AI 600-1) warns that organizations ship agents without role clarity, human-in-the-loop or incident playbooks. ✅ Model safety reduces bad text. Agent safety prevents bad state. Treat the agent like a junior SRE - with narrow permissions, audited actions and clear escalation paths. If you're building agents, what's the one control you won't ship without? Also drop a comment if you'd like me to share a follow-up post on how to tackle each of these risks. #AgenticAI #AISafety #AIAgents #GenAI #CyberSecurity #MCP #OWASP #NIST #AITrust #ResponsibleAI #AIForBusiness

5

Felt cute, so I added 64-bit support for WASI preview1 in Wasmtime, the de facto standard WebAssembly runtime. https://lnkd.in/gtHxr34X Key points: 1. You can now compile a C program to wasm64-wasip1 using my fork of wasi-sdk (https://lnkd.in/gwqKX8NE) and run the resulting binary. 2. This feature is not standards-compliant so it will not be upstreamed. The WASI team is unlikely to agree to a memory64 preview1 ABI because they are focusing on preview2 and beyond. 3. With that said, this is useful if you want to run non-production workloads (e.g. for research purpose or to understand how memory64 can impact your application performance) High level steps: 1. Download the modified wasi-sdk toolchain (https://lnkd.in/g2YJt4Ys) 2. Compile your program targeting wasm64-wasip1 using the toolchain 3. Clone my fork of Wasmtime and checkout the yagehu/wasip1-memory64 branch 4. Build it with `cargo build --release` 5. Now you can run the wasm64 binary with `wasmtime run --wasm memory64 binary.wasm`

13

I am excited to share that our paper, "LEGO: A Layout Expression Language for Code Generation of Hierarchical Mapping," will be presented at CGO 2026 next week! In this work, we introduce LEGO, a new abstraction that decouples computation from layout details. Unlike previous approaches like CuTe and Graphene, LEGO: 1) Automated Stride Derivation: LEGO eliminates explicit stride specifications. Deriving index expressions directly from layout definitions removes the burden of manual low-level bookkeeping. 2) Flexible Integration for Tools: Built on Python and SymPy, LEGO provides a lightweight infrastructure that you can easily embed into your own tools and domain-specific languages (DSLs). 3) Versatile Code Generation: We demonstrate adaptability by targeting diverse backends like Triton and CUDA, and introduce a custom SymPy to MLIR backend to facilitate integration with modern compiler frameworks. 4) Generalized Bijective Support: Moving beyond rectangular shape algebras, LEGO supports any bijective mapping, including user-defined functions. This enables complex patterns, such as antidiagonal layouts. Paper: https://lnkd.in/gwFkTRsE Code: https://lnkd.in/gRvTrdBN Big thanks to my co-authors Cosmin E. Oancea and Mary Hall!

160

11 Comments

🧠 AI Rights Historical Wrongs Researchers at Stanford and Princeton have fine-tuned a large language model to uncover racially discriminatory housing clauses buried in millions of historical property deeds — accelerating a process that could have taken a decade by hand. 📍 The problem: Many property deeds in Northern California — especially in Santa Clara County — still contain illegal racial covenants from as far back as the 1850s, banning people of color from owning or living in certain homes. 🔎 The solution: The team used OCR to digitize 5.2 million deed pages, then fine-tuned Mistral-7B via LoRA on a hand-labeled dataset of discriminatory phrases. The model learned to identify harmful language in context, avoiding false positives like the surnames "Black" or "White." 📊 Results: Detected 24,500 lots with racial clauses (~25% of homes in 1950) Found that 10 developers were responsible for 1/3 of all discriminatory clauses Model reviewed all pages in 6 days at a cost of $258 Equivalent manual effort: 10 years, $1.4M 💥 This work doesn't just use LLMs to build the future — it uses them to understand and repair the past. The open-sourced model now enables other U.S. counties to follow suit. #AI #MachineLearning #LLM #ResponsibleAI #Stanford #Princeton #HistoryMeetsAI #TechForGood #EthicalAI #SocialImpact #PropertyTech https://lnkd.in/dtec-zbx

4

Want to train LLM to reason right on your mac? Check out my article on Reinforcement Fine-Tuning LLMs with GRPO on Apple silicon using MLX framework. I remembered vividly when DeepSeek-R1 burst onto the scene this year, it dominated discussions over Chinese reunion dinners, AI stocks crashed, and friends I knew (including me) were busy asking it to reason their Bazi. A key idea introduced along with Deepseek R1 was Group Relative Policy Optimisation (GRPO), which helps to incentivise reasoning capability in LLMs via reinforcement learning. Since then, there have been courses such as Predibase's course on implementing GRPO, and improvements to the algorithm such as Mistral AI's Magistral LLM which was trained on a more efficient GRPO algorithm. Following the course by Predibase hosted on DeepLearning.AI, I have written an article where I explain step by step each component of GRPO equation, with a worked example to illustrate the algorithm. If you have on hand an Apple silicon computer, I have also included a Jupyter notebook which implements GRPO using Apple's MLX framework. Credits to Awni Hannun for creating the base notebook for LoRA Fine-Tuning with MLX LM. Full disclosure that I created the article and the notebook with the assistance of Google Gemini. If you noticed any errors with either the article or the notebook, please let me know or file a pull request.

889

39 Comments