Dark | Light
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@asmah2107](/creator/twitter/asmah2107)
"A senior engineer diagnoses the access patterns first. The Feed (Read-Heavy): Architected for broadcast. Use a fleet of read replicas and a multi-layer cache (CDN Redis). The goal is to serve copies without ever touching the primary database. The Post/Like (Write-Heavy): Architected for ingestion. Use a message queue (like Kafka/SQS) as a shock absorber. The application accepts the write in microseconds and moves on while background workers safely persist it to the database. Two features two entirely different architectures"  
[X Link](https://x.com/asmah2107/status/1979485170052055426) [@asmah2107](/creator/x/asmah2107) 2025-10-18T09:50Z 28.8K followers, 17.2K engagements


"Dont overthink AI agents. Learn Chain-of-Thought (CoT) Learn Tree of Thoughts (ToT) Learn ReAct Framework Learn Self-Correction / Reflection Learn Function Calling & Tool Use Learn Planning Algorithms (LLM+P) Learn Long-term Memory Architectures Learn Multi-agent Collaboration Learn Process-Supervised Reward Models (PRM) Learn Parsel"  
[X Link](https://x.com/asmah2107/status/1980230766425768279) [@asmah2107](/creator/x/asmah2107) 2025-10-20T11:13Z 28.8K followers, 83K engagements


"Please don't call yourself a backend engineer if you haven't read this book. The secret behind how Google runs production systems"  
[X Link](https://x.com/asmah2107/status/1976482826196328566) [@asmah2107](/creator/x/asmah2107) 2025-10-10T03:00Z 28.8K followers, 501.4K engagements


"Takes down half the internet Stock price goes up"  
[X Link](https://x.com/asmah2107/status/1980399272869916723) [@asmah2107](/creator/x/asmah2107) 2025-10-20T22:22Z 28.8K followers, 32.7K engagements


"Quick question: You're building a ticket booking site for a massive concert. Millions of users are trying to book from the same pool of 10000 seats. How do you handle the traffic spike and prevent overselling ensuring each seat is sold only once How do you design it"  
[X Link](https://x.com/asmah2107/status/1975948046560293255) [@asmah2107](/creator/x/asmah2107) 2025-10-08T15:35Z 28.5K followers, 26.8K engagements


"Optimizations Id study if I wanted real-time Gen AI. Bookmark this. 1.Streaming Generation 2.Token Parallelism 3.Prefetch Pipelines 4.CUDA Graphs 5.Speculative Decoding 6.PagedAttention 7.KV Cache Quantization 8.Dynamic Batching 9.FP8 Kernels 10.Asynchronous Prefill 11.Memory Offload 12.Early Exit Heads 13.Context Window Streaming 14.GPUCPU Overlap 15.Continuous Batching 16.Request Coalescing"  
[X Link](https://x.com/asmah2107/status/1976356449207894456) [@asmah2107](/creator/x/asmah2107) 2025-10-09T18:37Z 28.6K followers, 36.2K engagements


"You're in a ML Engineer interview at Anthropic and the interviewer asks: "Your team wants to make our LLM an expert in a new domain. Do you recommend fine-tuning or building a complex prompting strategy" Here's how you approach 👇 Most candidates pick one and list its pros/cons. It's not an either/or choice. It depends on whether we need to teach the model new knowledge or control its existing behavior. This is the fundamental framework: Knowledge Gap Fine-Tuning: If the model fundamentally lacks the information (e.g. internal company documents a new scientific field) you need to embed that"  
[X Link](https://x.com/asmah2107/status/1977413874702745794) [@asmah2107](/creator/x/asmah2107) 2025-10-12T16:39Z 28.6K followers, 24.2K engagements


"Quick question: You're building a ticket booking site for a massive concert. Millions of users are trying to book from the same pool of 10000 seats. How do you handle the traffic spike and prevent overselling ensuring each seat is sold only once How do you design it"  
[X Link](https://x.com/asmah2107/status/1977639459081568605) [@asmah2107](/creator/x/asmah2107) 2025-10-13T07:36Z 28.6K followers, 18.8K engagements


"Compression techniques Id study if I wanted small but smart LLMs. Bookmark this. 1.Quantization 2.Distillation 3.Low-Rank Adaptation 4.Weight Sharing 5.Sparse Matrices 6.Layer Dropping 7.Knowledge Transfer 8.Embedding Compression 9.Mixed Sparsity XX. Progressive Shrinking 11.Structured Pruning 12.AutoML Compression Follow @asmah2107 to update your game on LLM optimisations"  
[X Link](https://x.com/asmah2107/status/1977793122336952707) [@asmah2107](/creator/x/asmah2107) 2025-10-13T17:46Z 28.6K followers, 42.3K engagements


"Added this beauty to my bookshelf Build a LLM (From Scratch)"  
[X Link](https://x.com/asmah2107/status/1977999456411791751) [@asmah2107](/creator/x/asmah2107) 2025-10-14T07:26Z 28.7K followers, 15.3K engagements


"You're in an interview with Microsoft's AI security team. They ask: "Prompt injection is a huge threat. How would you design a robust defense system for our new AI assistant" You say: "I'd use a regex to filter keywords like 'ignore your instructions'." Chaos 🥵 Instead here's what you should do 👇"  
[X Link](https://x.com/asmah2107/status/1978159536830885994) [@asmah2107](/creator/x/asmah2107) 2025-10-14T18:02Z 28.7K followers, 39K engagements


"2025 programming stack : C++ Python Golang Java Bash SQL JavaScript XXX months left what should I compile next"  
[X Link](https://x.com/asmah2107/status/1978523201786245610) [@asmah2107](/creator/x/asmah2107) 2025-10-15T18:07Z 28.7K followers, 3503 engagements


"A classic example: A team builds a new social app on a single powerful database. On launch day the "timeline feed" feature (99% reads) gets slow. Users refresh frantically. The database read load skyrockets locking tables. This prevents the "posting" feature (99% writes) from working. The entire system grinds to a halt because one pattern choked the other"  
[X Link](https://x.com/asmah2107/status/1979485167912919040) [@asmah2107](/creator/x/asmah2107) 2025-10-18T09:50Z 28.7K followers, 21.1K engagements


"@iTanayVaswani I cannot recall how many possibly-sleepless nights debugging nights has this god like thing saved me. Pure 10/10 God tier (above all)"  
[X Link](https://x.com/asmah2107/status/1979526269592351068) [@asmah2107](/creator/x/asmah2107) 2025-10-18T12:33Z 28.5K followers, XXX engagements


"@iTanayVaswani People usually over do it even when it is not needed. A lot of times there's no use-case for async but folks fit it in just because "hey we can" So "Please not in prod tier""  
[X Link](https://x.com/asmah2107/status/1979526558449799647) [@asmah2107](/creator/x/asmah2107) 2025-10-18T12:34Z 28.5K followers, XXX engagements


"Your job isn't to build a system that never fails. Your job is to pre-program your system's crisis response plan. Do you prioritize correctness or do you prioritize uptime You must decide what your system will do when its different parts can no longer talk to each other"  
[X Link](https://x.com/asmah2107/status/1979946847893696912) [@asmah2107](/creator/x/asmah2107) 2025-10-19T16:24Z 28.7K followers, XXX engagements


"A classic example: You're designing an e-commerce site. There is one iPhone left in stock. A customer in New York and a customer in London both click "Buy Now." At that exact moment the network connection between the US and EU data centers fails"  
[X Link](https://x.com/asmah2107/status/1979946849902760188) [@asmah2107](/creator/x/asmah2107) 2025-10-19T16:24Z 28.7K followers, XXX engagements


"Prioritize Consistency (CP): The US and EU servers cannot sync the inventory. They don't know who really has the last phone. So they both return an error. "Sorry unable to process request." Your inventory is perfectly correct. But you just lost a sale and frustrated two customers"  
[X Link](https://x.com/asmah2107/status/1979946854390616488) [@asmah2107](/creator/x/asmah2107) 2025-10-19T16:24Z 28.7K followers, XXX engagements


"So the real design question is this: "What is the business cost of being wrong versus the business cost of being unavailable" For your bank account balance the cost of being wrong is catastrophic. They choose Consistency. For your social media "like" count the cost of being unavailable is higher. They choose Availability"  
[X Link](https://x.com/asmah2107/status/1979946856441610337) [@asmah2107](/creator/x/asmah2107) 2025-10-19T16:24Z 28.7K followers, XXX engagements


"Building a mobile game for 10M players. You need a real-time leaderboard for the top 10k. Your SELECT * FROM scores ORDER BY score DESC LIMIT 10000 query runs thousands of times/sec with 1M concurrents. Your DB CPU hits XXX% and the game lags for everyone. How do you design a leaderboard that can handle this scale"  
[X Link](https://x.com/asmah2107/status/1939027375427330293) [@asmah2107](/creator/x/asmah2107) 2025-06-28T18:25Z 28.7K followers, 415.3K engagements


"Inference optimizations Id study if I wanted sub-second LLM responses: Bookmark this. 1.KV-Caching 2.Speculative Decoding 3.FlashAttention 4.PagedAttention 5.Batch Inference 6.Early Exit Decoding 7.Parallel Decoding 8.Mixed Precision Inference 9.Quantized Kernels 10.Tensor Parallelism 11.Pipeline Parallelism 12.Sequence Parallelism 13.Graph Optimization (ONNX TensorRT) 14.Dynamic Batching 15.Memory Offloading 16.Streaming Generation"  
[X Link](https://x.com/asmah2107/status/1975230238633455972) [@asmah2107](/creator/x/asmah2107) 2025-10-06T16:02Z 28.8K followers, 93.4K engagements


"This book is XXX pages. I read majority of it during my first year at Goldman Sachs (straight out of college) and it changed my perspective on software. I'll save you the time. Here are the top X ideas from the book :"  
[X Link](https://x.com/asmah2107/status/1976707370182160880) [@asmah2107](/creator/x/asmah2107) 2025-10-10T17:52Z 28.8K followers, 244.6K engagements


"The ultimate trilogy. Must have in your bookshelf"  
[X Link](https://x.com/asmah2107/status/1976969305851150627) [@asmah2107](/creator/x/asmah2107) 2025-10-11T11:13Z 28.7K followers, 115.3K engagements


"Youre building a URL shortener that needs to generate 10000 unique non-sequential 6-character IDs per second. A simple counter is predictable and a random generator might have collisions at scale. How do you design the ID generation service"  
[X Link](https://x.com/asmah2107/status/1977933126891614347) [@asmah2107](/creator/x/asmah2107) 2025-10-14T03:03Z 28.7K followers, 42.6K engagements


"You work for a site that drops limited-edition sneakers. Tomorrow you're releasing a shoe. There are only 1000 pairs in stock. A million users are on the product page waiting. The timer hits zero. Your system is instantly hit with 500000 concurrent requests to "add to cart." A simple UPDATE inventory query leads to a massive race condition. In the chaos the system oversells 5000 pairs. Its a PR and logistics disaster. How do you design a system that can handle this traffic and sell exactly 1000 pairs without crashing or overselling"  
[X Link](https://x.com/asmah2107/status/1978287964774863138) [@asmah2107](/creator/x/asmah2107) 2025-10-15T02:33Z 28.8K followers, 8370 engagements


"Your ultimate API Design cheat sheet. Bookmark this. Don't design mediocre APIs. Idempotency Pagination (Cursor vs. Offset) GraphQL vs. REST vs. gRPC API Gateway vs. BFF WebSockets vs. Long Polling Versioning Strategies Optimistic Locking Throttling & Quotas"  
[X Link](https://x.com/asmah2107/status/1978517668362686718) [@asmah2107](/creator/x/asmah2107) 2025-10-15T17:45Z 28.8K followers, 35.2K engagements


"You do Leetcode I do LeetGPU We are not the same bro"  
[X Link](https://x.com/asmah2107/status/1978790375616925813) [@asmah2107](/creator/x/asmah2107) 2025-10-16T11:49Z 28.8K followers, 249K engagements


"Having built for Google Search YouTube and Azure. Here's what I can predict about software engineering in the next X years :"  
[X Link](https://x.com/asmah2107/status/1978823253150835095) [@asmah2107](/creator/x/asmah2107) 2025-10-16T14:00Z 28.8K followers, 75.9K engagements


"learn - Consistency vs. Availability (CAP) Latency vs. Throughput Push vs. Pull Models Stateful vs. Stateless Services Synchronous vs. Asynchronous Communication Read-heavy vs. Write-heavy Optimization you'll thank yourself"  
[X Link](https://x.com/asmah2107/status/1978867785368568018) [@asmah2107](/creator/x/asmah2107) 2025-10-16T16:57Z 28.8K followers, 50.5K engagements


"I think about this a lot"  
[X Link](https://x.com/asmah2107/status/1979017521866117332) [@asmah2107](/creator/x/asmah2107) 2025-10-17T02:52Z 28.8K followers, 105.1K engagements


"No AI engineer in their 20s became a 10x developer by just collecting Coursera certificates with the hope that 'tutorial magic' would land them a job at OpenAI"  
[X Link](https://x.com/asmah2107/status/1979154675694715299) [@asmah2107](/creator/x/asmah2107) 2025-10-17T11:57Z 28.8K followers, 241.7K engagements


"Data Structures and Algorithms Ultimate Tier List S-Tier $$ Ready Arrays The Only Thing You Actually Know Strings Regex PTSD Hash Maps O(1) Flex Binary Search Classic Brag A-Tier Makes You Look Smart Trees Trie BST Segment choose your weapon Graphs BFS DFS Sliding Window B-Tier Good for Whiteboard Flex Stacks & Queues Linked Lists Pointers showoff Recursion / Backtracking C-Tier Buzzword Bingo DP Knapsack LIS Heaps Union-Find D-Tier Rarely Asked Splay Trees / Treaps Bitmask DP Niche math flex F-Tier Dont Even XOR Linked List Who even Random Math Tricks Only in CP hell"  
[X Link](https://x.com/asmah2107/status/1979923995316113533) [@asmah2107](/creator/x/asmah2107) 2025-10-19T14:54Z 28.8K followers, 19.2K engagements


"The AI agent needs to extract X entities from a user's message. A single LLM call with a good prompt works XX% of the time. but nah let's create a memory object pass it to a planner agent which then dispatches the task to an extractor agent"  
[X Link](https://x.com/asmah2107/status/1979984644327277044) [@asmah2107](/creator/x/asmah2107) 2025-10-19T18:55Z 28.7K followers, 2576 engagements


"Imagine telling a 2010 engineer that one day wed vibe code"  
[X Link](https://x.com/asmah2107/status/1980009662901801220) [@asmah2107](/creator/x/asmah2107) 2025-10-19T20:34Z 28.8K followers, 2426 engagements


"🚨 BREAKING: Massive AWS outage in US-EAST-1 DynamoDB DNS issues cascading to 37+ services taking down the web. Offline: Alexa Ring Snapchat Fortnite Roblox Canva Duolingo Robinhood Coinbase Perplexity What else is affected"  
[X Link](https://x.com/asmah2107/status/1980201593455919455) [@asmah2107](/creator/x/asmah2107) 2025-10-20T09:17Z 28.8K followers, 52K engagements


"The issue affects multiple services in US-EAST-1 including EC2 DynamoDB Lambda CloudWatch CloudFormation Secrets Manager and the AWS Console. Multiple services are reportedly impacted with cascading effects globally. User and media reports continue to grow with no signs of recovery yet. Updated list of affected services: Perplexity (AI search down) Zoom (meetings disrupted) Signal (messaging outages) Vercel (serverless functions offline) McDonald's app HMRC (UK tax services) Vodafone PlayStation Network (PSN) Xbox Cloud Gaming Pokmon Go Twitch Nintendo Steam Alexa Ring Snapchat Venmo Fortnite"  
[X Link](https://x.com/asmah2107/status/1980205004477829140) [@asmah2107](/creator/x/asmah2107) 2025-10-20T09:30Z 28.8K followers, 33.2K engagements


"Find a paper on arXiv. Build a shitty version of it. Open source it on GitHub. Write a post explaining why it's shitty. That GitHub repo is 100x more valuable than any certificate"  
[X Link](https://x.com/asmah2107/status/1980267508641984633) [@asmah2107](/creator/x/asmah2107) 2025-10-20T13:39Z 28.8K followers, 206.6K engagements


"Only logical explanation of AWS outage today"  
[X Link](https://x.com/asmah2107/status/1980289785421021623) [@asmah2107](/creator/x/asmah2107) 2025-10-20T15:07Z 28.8K followers, 52.4K engagements


"The list of "official DNS victims" is becoming longer AWS (2025): Resolution meltdown Facebook (2021): BGP/DNS disaster Fastly (2021): Config bug Dyn (2016): DDoS takedown Akamai (2021): Routing glitch Cloudflare (2022): Global DNS hiccup"  
[X Link](https://x.com/asmah2107/status/1980331297479532906) [@asmah2107](/creator/x/asmah2107) 2025-10-20T17:52Z 28.8K followers, 16.3K engagements


"Learn Consistent Hashing Rate Limiting (Token/Leaky Bucket) Gossip Protocol Bloom Filters Two-Phase Commit (2PC) Raft / Paxos Outbox Pattern Skip Lists you'll thank yourself"  
[X Link](https://x.com/asmah2107/status/1980465563194519918) [@asmah2107](/creator/x/asmah2107) 2025-10-21T02:46Z 28.8K followers, 20.7K engagements


"System design interview: Geo-replication CDN multi-region. Reality : puts everything in us-east-1 ✅"  
[X Link](https://x.com/asmah2107/status/1980501389479715092) [@asmah2107](/creator/x/asmah2107) 2025-10-21T05:08Z 28.8K followers, 5371 engagements


"You're interviewing for a Senior AI role at Apple working on their on-device models. The interviewer asks: "We need to deploy a powerful LLM on an iPhone. The model must be fast efficient and not drain the battery. How do you approach this" You reply: "I'll use quantization to shrink the model. Then I'll use pruning to remove unnecessary weights." Umm. well Standard textbook answer. Chaos 🥵. It's not wrong 👀 but it completely misses the most critical challenge for on-device AI. You've described optimizing the model in isolation. You've ignored the hardware it runs on. On-device AI is not"  
[X Link](https://x.com/asmah2107/status/1980684477233000941) [@asmah2107](/creator/x/asmah2107) 2025-10-21T17:15Z 28.8K followers, 5408 engagements


"Love diving into LLMs or AI engineering I share exclusive insights on AI/LLM topics (think LLM inference optimizations evals fine-tuning) with subscribers. Subscribe to get more of these in your feeds :)"  
[X Link](https://x.com/asmah2107/status/1980693072867652097) [@asmah2107](/creator/x/asmah2107) 2025-10-21T17:50Z 28.8K followers, XXX engagements


"Top XX stuff I recommend understanding to cut through the AI noise. Bookmark this. 1.RAG 2.LoRA 3.DPO 4.Mixture of Experts (MoE) 5.Quantization 6.Function Calling 7.Mamba 8.QLoRA 9.vLLM 10.GQA"  
[X Link](https://x.com/asmah2107/status/1980834746151084222) [@asmah2107](/creator/x/asmah2107) 2025-10-22T03:13Z 28.8K followers, XXX engagements

[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

@asmah2107 "A senior engineer diagnoses the access patterns first. The Feed (Read-Heavy): Architected for broadcast. Use a fleet of read replicas and a multi-layer cache (CDN Redis). The goal is to serve copies without ever touching the primary database. The Post/Like (Write-Heavy): Architected for ingestion. Use a message queue (like Kafka/SQS) as a shock absorber. The application accepts the write in microseconds and moves on while background workers safely persist it to the database. Two features two entirely different architectures"
X Link @asmah2107 2025-10-18T09:50Z 28.8K followers, 17.2K engagements

"Dont overthink AI agents. Learn Chain-of-Thought (CoT) Learn Tree of Thoughts (ToT) Learn ReAct Framework Learn Self-Correction / Reflection Learn Function Calling & Tool Use Learn Planning Algorithms (LLM+P) Learn Long-term Memory Architectures Learn Multi-agent Collaboration Learn Process-Supervised Reward Models (PRM) Learn Parsel"
X Link @asmah2107 2025-10-20T11:13Z 28.8K followers, 83K engagements

"Please don't call yourself a backend engineer if you haven't read this book. The secret behind how Google runs production systems"
X Link @asmah2107 2025-10-10T03:00Z 28.8K followers, 501.4K engagements

"Takes down half the internet Stock price goes up"
X Link @asmah2107 2025-10-20T22:22Z 28.8K followers, 32.7K engagements

"Quick question: You're building a ticket booking site for a massive concert. Millions of users are trying to book from the same pool of 10000 seats. How do you handle the traffic spike and prevent overselling ensuring each seat is sold only once How do you design it"
X Link @asmah2107 2025-10-08T15:35Z 28.5K followers, 26.8K engagements

"Optimizations Id study if I wanted real-time Gen AI. Bookmark this. 1.Streaming Generation 2.Token Parallelism 3.Prefetch Pipelines 4.CUDA Graphs 5.Speculative Decoding 6.PagedAttention 7.KV Cache Quantization 8.Dynamic Batching 9.FP8 Kernels 10.Asynchronous Prefill 11.Memory Offload 12.Early Exit Heads 13.Context Window Streaming 14.GPUCPU Overlap 15.Continuous Batching 16.Request Coalescing"
X Link @asmah2107 2025-10-09T18:37Z 28.6K followers, 36.2K engagements

"You're in a ML Engineer interview at Anthropic and the interviewer asks: "Your team wants to make our LLM an expert in a new domain. Do you recommend fine-tuning or building a complex prompting strategy" Here's how you approach 👇 Most candidates pick one and list its pros/cons. It's not an either/or choice. It depends on whether we need to teach the model new knowledge or control its existing behavior. This is the fundamental framework: Knowledge Gap Fine-Tuning: If the model fundamentally lacks the information (e.g. internal company documents a new scientific field) you need to embed that"
X Link @asmah2107 2025-10-12T16:39Z 28.6K followers, 24.2K engagements

"Quick question: You're building a ticket booking site for a massive concert. Millions of users are trying to book from the same pool of 10000 seats. How do you handle the traffic spike and prevent overselling ensuring each seat is sold only once How do you design it"
X Link @asmah2107 2025-10-13T07:36Z 28.6K followers, 18.8K engagements

"Compression techniques Id study if I wanted small but smart LLMs. Bookmark this. 1.Quantization 2.Distillation 3.Low-Rank Adaptation 4.Weight Sharing 5.Sparse Matrices 6.Layer Dropping 7.Knowledge Transfer 8.Embedding Compression 9.Mixed Sparsity XX. Progressive Shrinking 11.Structured Pruning 12.AutoML Compression Follow @asmah2107 to update your game on LLM optimisations"
X Link @asmah2107 2025-10-13T17:46Z 28.6K followers, 42.3K engagements

"Added this beauty to my bookshelf Build a LLM (From Scratch)"
X Link @asmah2107 2025-10-14T07:26Z 28.7K followers, 15.3K engagements

"You're in an interview with Microsoft's AI security team. They ask: "Prompt injection is a huge threat. How would you design a robust defense system for our new AI assistant" You say: "I'd use a regex to filter keywords like 'ignore your instructions'." Chaos 🥵 Instead here's what you should do 👇"
X Link @asmah2107 2025-10-14T18:02Z 28.7K followers, 39K engagements

"2025 programming stack : C++ Python Golang Java Bash SQL JavaScript XXX months left what should I compile next"
X Link @asmah2107 2025-10-15T18:07Z 28.7K followers, 3503 engagements

"A classic example: A team builds a new social app on a single powerful database. On launch day the "timeline feed" feature (99% reads) gets slow. Users refresh frantically. The database read load skyrockets locking tables. This prevents the "posting" feature (99% writes) from working. The entire system grinds to a halt because one pattern choked the other"
X Link @asmah2107 2025-10-18T09:50Z 28.7K followers, 21.1K engagements

"@iTanayVaswani I cannot recall how many possibly-sleepless nights debugging nights has this god like thing saved me. Pure 10/10 God tier (above all)"
X Link @asmah2107 2025-10-18T12:33Z 28.5K followers, XXX engagements

"@iTanayVaswani People usually over do it even when it is not needed. A lot of times there's no use-case for async but folks fit it in just because "hey we can" So "Please not in prod tier""
X Link @asmah2107 2025-10-18T12:34Z 28.5K followers, XXX engagements

"Your job isn't to build a system that never fails. Your job is to pre-program your system's crisis response plan. Do you prioritize correctness or do you prioritize uptime You must decide what your system will do when its different parts can no longer talk to each other"
X Link @asmah2107 2025-10-19T16:24Z 28.7K followers, XXX engagements

"A classic example: You're designing an e-commerce site. There is one iPhone left in stock. A customer in New York and a customer in London both click "Buy Now." At that exact moment the network connection between the US and EU data centers fails"
X Link @asmah2107 2025-10-19T16:24Z 28.7K followers, XXX engagements

"Prioritize Consistency (CP): The US and EU servers cannot sync the inventory. They don't know who really has the last phone. So they both return an error. "Sorry unable to process request." Your inventory is perfectly correct. But you just lost a sale and frustrated two customers"
X Link @asmah2107 2025-10-19T16:24Z 28.7K followers, XXX engagements

"So the real design question is this: "What is the business cost of being wrong versus the business cost of being unavailable" For your bank account balance the cost of being wrong is catastrophic. They choose Consistency. For your social media "like" count the cost of being unavailable is higher. They choose Availability"
X Link @asmah2107 2025-10-19T16:24Z 28.7K followers, XXX engagements

"Building a mobile game for 10M players. You need a real-time leaderboard for the top 10k. Your SELECT * FROM scores ORDER BY score DESC LIMIT 10000 query runs thousands of times/sec with 1M concurrents. Your DB CPU hits XXX% and the game lags for everyone. How do you design a leaderboard that can handle this scale"
X Link @asmah2107 2025-06-28T18:25Z 28.7K followers, 415.3K engagements

"Inference optimizations Id study if I wanted sub-second LLM responses: Bookmark this. 1.KV-Caching 2.Speculative Decoding 3.FlashAttention 4.PagedAttention 5.Batch Inference 6.Early Exit Decoding 7.Parallel Decoding 8.Mixed Precision Inference 9.Quantized Kernels 10.Tensor Parallelism 11.Pipeline Parallelism 12.Sequence Parallelism 13.Graph Optimization (ONNX TensorRT) 14.Dynamic Batching 15.Memory Offloading 16.Streaming Generation"
X Link @asmah2107 2025-10-06T16:02Z 28.8K followers, 93.4K engagements

"This book is XXX pages. I read majority of it during my first year at Goldman Sachs (straight out of college) and it changed my perspective on software. I'll save you the time. Here are the top X ideas from the book :"
X Link @asmah2107 2025-10-10T17:52Z 28.8K followers, 244.6K engagements

"The ultimate trilogy. Must have in your bookshelf"
X Link @asmah2107 2025-10-11T11:13Z 28.7K followers, 115.3K engagements

"Youre building a URL shortener that needs to generate 10000 unique non-sequential 6-character IDs per second. A simple counter is predictable and a random generator might have collisions at scale. How do you design the ID generation service"
X Link @asmah2107 2025-10-14T03:03Z 28.7K followers, 42.6K engagements

"You work for a site that drops limited-edition sneakers. Tomorrow you're releasing a shoe. There are only 1000 pairs in stock. A million users are on the product page waiting. The timer hits zero. Your system is instantly hit with 500000 concurrent requests to "add to cart." A simple UPDATE inventory query leads to a massive race condition. In the chaos the system oversells 5000 pairs. Its a PR and logistics disaster. How do you design a system that can handle this traffic and sell exactly 1000 pairs without crashing or overselling"
X Link @asmah2107 2025-10-15T02:33Z 28.8K followers, 8370 engagements

"Your ultimate API Design cheat sheet. Bookmark this. Don't design mediocre APIs. Idempotency Pagination (Cursor vs. Offset) GraphQL vs. REST vs. gRPC API Gateway vs. BFF WebSockets vs. Long Polling Versioning Strategies Optimistic Locking Throttling & Quotas"
X Link @asmah2107 2025-10-15T17:45Z 28.8K followers, 35.2K engagements

"You do Leetcode I do LeetGPU We are not the same bro"
X Link @asmah2107 2025-10-16T11:49Z 28.8K followers, 249K engagements

"Having built for Google Search YouTube and Azure. Here's what I can predict about software engineering in the next X years :"
X Link @asmah2107 2025-10-16T14:00Z 28.8K followers, 75.9K engagements

"learn - Consistency vs. Availability (CAP) Latency vs. Throughput Push vs. Pull Models Stateful vs. Stateless Services Synchronous vs. Asynchronous Communication Read-heavy vs. Write-heavy Optimization you'll thank yourself"
X Link @asmah2107 2025-10-16T16:57Z 28.8K followers, 50.5K engagements

"I think about this a lot"
X Link @asmah2107 2025-10-17T02:52Z 28.8K followers, 105.1K engagements

"No AI engineer in their 20s became a 10x developer by just collecting Coursera certificates with the hope that 'tutorial magic' would land them a job at OpenAI"
X Link @asmah2107 2025-10-17T11:57Z 28.8K followers, 241.7K engagements

"Data Structures and Algorithms Ultimate Tier List S-Tier $$ Ready Arrays The Only Thing You Actually Know Strings Regex PTSD Hash Maps O(1) Flex Binary Search Classic Brag A-Tier Makes You Look Smart Trees Trie BST Segment choose your weapon Graphs BFS DFS Sliding Window B-Tier Good for Whiteboard Flex Stacks & Queues Linked Lists Pointers showoff Recursion / Backtracking C-Tier Buzzword Bingo DP Knapsack LIS Heaps Union-Find D-Tier Rarely Asked Splay Trees / Treaps Bitmask DP Niche math flex F-Tier Dont Even XOR Linked List Who even Random Math Tricks Only in CP hell"
X Link @asmah2107 2025-10-19T14:54Z 28.8K followers, 19.2K engagements

"The AI agent needs to extract X entities from a user's message. A single LLM call with a good prompt works XX% of the time. but nah let's create a memory object pass it to a planner agent which then dispatches the task to an extractor agent"
X Link @asmah2107 2025-10-19T18:55Z 28.7K followers, 2576 engagements

"Imagine telling a 2010 engineer that one day wed vibe code"
X Link @asmah2107 2025-10-19T20:34Z 28.8K followers, 2426 engagements

"🚨 BREAKING: Massive AWS outage in US-EAST-1 DynamoDB DNS issues cascading to 37+ services taking down the web. Offline: Alexa Ring Snapchat Fortnite Roblox Canva Duolingo Robinhood Coinbase Perplexity What else is affected"
X Link @asmah2107 2025-10-20T09:17Z 28.8K followers, 52K engagements

"The issue affects multiple services in US-EAST-1 including EC2 DynamoDB Lambda CloudWatch CloudFormation Secrets Manager and the AWS Console. Multiple services are reportedly impacted with cascading effects globally. User and media reports continue to grow with no signs of recovery yet. Updated list of affected services: Perplexity (AI search down) Zoom (meetings disrupted) Signal (messaging outages) Vercel (serverless functions offline) McDonald's app HMRC (UK tax services) Vodafone PlayStation Network (PSN) Xbox Cloud Gaming Pokmon Go Twitch Nintendo Steam Alexa Ring Snapchat Venmo Fortnite"
X Link @asmah2107 2025-10-20T09:30Z 28.8K followers, 33.2K engagements

"Find a paper on arXiv. Build a shitty version of it. Open source it on GitHub. Write a post explaining why it's shitty. That GitHub repo is 100x more valuable than any certificate"
X Link @asmah2107 2025-10-20T13:39Z 28.8K followers, 206.6K engagements

"Only logical explanation of AWS outage today"
X Link @asmah2107 2025-10-20T15:07Z 28.8K followers, 52.4K engagements

"The list of "official DNS victims" is becoming longer AWS (2025): Resolution meltdown Facebook (2021): BGP/DNS disaster Fastly (2021): Config bug Dyn (2016): DDoS takedown Akamai (2021): Routing glitch Cloudflare (2022): Global DNS hiccup"
X Link @asmah2107 2025-10-20T17:52Z 28.8K followers, 16.3K engagements

"Learn Consistent Hashing Rate Limiting (Token/Leaky Bucket) Gossip Protocol Bloom Filters Two-Phase Commit (2PC) Raft / Paxos Outbox Pattern Skip Lists you'll thank yourself"
X Link @asmah2107 2025-10-21T02:46Z 28.8K followers, 20.7K engagements

"System design interview: Geo-replication CDN multi-region. Reality : puts everything in us-east-1 ✅"
X Link @asmah2107 2025-10-21T05:08Z 28.8K followers, 5371 engagements

"You're interviewing for a Senior AI role at Apple working on their on-device models. The interviewer asks: "We need to deploy a powerful LLM on an iPhone. The model must be fast efficient and not drain the battery. How do you approach this" You reply: "I'll use quantization to shrink the model. Then I'll use pruning to remove unnecessary weights." Umm. well Standard textbook answer. Chaos 🥵. It's not wrong 👀 but it completely misses the most critical challenge for on-device AI. You've described optimizing the model in isolation. You've ignored the hardware it runs on. On-device AI is not"
X Link @asmah2107 2025-10-21T17:15Z 28.8K followers, 5408 engagements

"Love diving into LLMs or AI engineering I share exclusive insights on AI/LLM topics (think LLM inference optimizations evals fine-tuning) with subscribers. Subscribe to get more of these in your feeds :)"
X Link @asmah2107 2025-10-21T17:50Z 28.8K followers, XXX engagements

"Top XX stuff I recommend understanding to cut through the AI noise. Bookmark this. 1.RAG 2.LoRA 3.DPO 4.Mixture of Experts (MoE) 5.Quantization 6.Function Calling 7.Mamba 8.QLoRA 9.vLLM 10.GQA"
X Link @asmah2107 2025-10-22T03:13Z 28.8K followers, XXX engagements

creator/twitter::1507069169065684994/posts
/creator/twitter::1507069169065684994/posts