ByteByteGo

⏳ LIMITED TIME OFFER: All in One Interview Prep Black Friday Sale Yearly Black Friday sale is now live! Use code BF2025 at checkout to get 30% off the all-in-one interview prep online courses. To take advantage of this limited time offer, subscribe before 11:59 pm PST on Monday, December 1. Get it here: https://bytebytego.com .

💬 How Uber Built a Conversational AI for Financial Analysis For Uber’s finance teams, getting answers from data once meant juggling multiple tools, writing SQL queries, and waiting hours for data requests. Today, it takes just one message in Slack. This is made possible by Finch - Uber’s conversational AI data agent. Finch is built to bring real-time financial intelligence directly into the daily workflow of analysts. When a user asks a question, Finch automatically identifies the right data source, generates a SQL query, checks access permissions, and delivers the answer instantly inside Slack. At its core, Finch is powered by: 1 - Curated financial data marts for reliable and structured data. 2 - A semantic layer built on OpenSearch that maps natural language to database columns for accurate query generation. 3 - An agentic workflow using LangChain and LangGraph, where specialized agents like the SQL Writer and Supervisor work together to understand, plan, and execute queries. 4 - Slack SDK and Google Sheets Explorer for seamless interaction and data sharing. By combining curated financial data, a semantic layer, and an agentic workflow, Finch delivers real-time analysis directly inside Slack. 🔗 Read the full breakdown: https://lnkd.in/e8BB8jSP Supported by our partners helping engineering teams build fast, reliable data systems that power real-time applications: Redpanda - a modern streaming platform designed for high-throughput, low-latency data pipelines. ➡️ https://bit.ly/4oXHjMH

🌐 How Grab Built an AI Foundational Model to Understand Customers Better Grab is no longer just a ride-hailing app. It is a superapp with food delivery, groceries, mobility, payments, and financial services. This creates an enormous stream of user interactions, but until recently, personalization relied heavily on manually engineered features that were brittle and slow to scale. Grab's engineering team changed it by building a foundation model that learns directly from user data to make sense of long-term user traits and short-term intent. The model handles text queries, numerical values, and geolocation data to support a personalization engine that powers ads, recommendations, fraud systems, and other retention models across the company. A few key points that stand out: 1 - Grab designed a transformer-based architecture that uses a key-value token format and modality-specific adapters. 2 - Training the model involves unsupervised pre-training with masked token prediction and next action prediction. 3 - Model supports two usage modes. Teams can fine-tune it for specific tasks like fraud detection, or they can extract embeddings as universal behavioral features for other models. This model is helping Grab build a scalable AI foundation for how superapps will understand customers better. By learning directly from user data instead of relying on brittle manual features, the model gives Grab a scalable foundation for personalization across the superapp. 🔗 Read the full breakdown: https://lnkd.in/eD_fguVB Supported by our partners building tools that help engineering teams maintain reliability as personalization systems grow in complexity: Sentry - AI Code Review that catches issues early and improves code quality as your codebase and models scale. ➡️ https://bit.ly/3M8t3SE

⚙️ How Datadog Built a Custom Database for Billions of Metrics Per Second Datadog’s monitoring platform handles an unimaginable scale of billions of data points flowing in every second from millions of servers. To keep up, Datadog built Monocle, a custom time-series database written in Rust - optimized for raw performance, reliability, and cost efficiency. Monocle runs on a thread-per-core model and uses an LSM-Tree storage design for extreme write throughput. At the architectural level, Datadog’s Metrics Platform splits data into two specialized systems: 1. A Long-Term Store for historical analytics 2. A Real-Time Store for live dashboards and alerts - serving 99% of queries Each incoming data point is first sent to Kafka, which powers data distribution across nodes, write-ahead logging for crash recovery, and automatic replication across availability zones for durability. Performance is maintained under heavy load through key systems: 1. Admission Control, which protects the cluster from overload 2. Cost-Based Scheduling, which prioritizes queries dynamically to maintain low latency 🔗 To see how these systems work together under real Datadog-scale load, read the full breakdown: https://lnkd.in/esMfUBPA Supported by our partners helping teams build and scale reliably: Datadog - Powering observability at scale. Download the On-Call Best Practices guide:  https://bit.ly/3Xxk0wZ SonarSource - Bridging the gap between AI-generated code and human-grade quality. Verify every line for security, maintainability, and trust: https://bit.ly/47VpU00

💡 What are AI Agents? Traditional software follows a predetermined path. However, AI agents can navigate uncertain situations and figure out what needs to be done. AI Agents can perceive, decide, and adapt to achieve goals. This represents a significant leap from static programs to dynamic collaborations. At its core, an AI agent works in a continuous cycle during which it perceives the current situation, thinks about what to do next, acts by taking a specific step, observes the results of the action, and then repeats the process. This cycle continues until the agent determines it has completed the task or needs human input to proceed further. Multiple types of AI Agents exist, each supporting different capabilities: 1 - Simple Reflect Agents react to patterns, like thermostats or basic chatbots. 2 - Model-Based Agents build internal maps of their environment, enabling context-aware behavior. 3 - Goal-Based Agents can plan ahead and choose actions that serve specific objectives. 4 - Utility-Based Agents weigh trade-offs to find the best possible outcome. 5 - Learning Agents improve continuously by learning from feedback and experience. AI agents are ushering in an era where software systems can become active collaborators. 🔗 Read the full breakdown: https://lnkd.in/eU3uG9EN Supported by our partners at You.com - helping companies unlock real ROI from AI. Download the full AI use case guide: https://bit.ly/4oQObem

🚀 Why Anthropic’s MCP Is a Big Deal AI models today can reason - but they can’t reliably act.  Once trained, they lose access to real-time data, APIs, and tools. Anthropic’s Model Context Protocol (MCP) aims to fix that. It’s a universal connector for AI - a shared standard that lets models talk to databases, APIs, and other systems safely. MCP standardizes communication between AI and external tools - much like how USB or HTTP unified earlier eras of computing. Three key parts: the Host (AI app), Client (translator), and Server (data bridge). Together, they let models access live information and take real-world actions. Open by design: Anthropic released MCP as open source to encourage broad adoption and prevent fragmentation across AI systems. Why it matters: This could turn today’s static models into connected agents that interact with live data - powering the next generation of AI assistants and workflows. 🔗 Read the full breakdown: https://lnkd.in/e_dXj28N Supported by our partners building tools that help developers move faster: Sentry - AI Code Review that catches issues before they ship: http://bit.ly/43w8iGo Conduktor - Helping teams avoid costly Kafka pitfalls at scale: https://bit.ly/4qCz5Ll

10 Key Data Structures We Use Every Day  . . - list: keep your Twitter feeds  - stack: support undo/redo of the word editor  - queue: keep printer jobs, or send user actions in-game  - hash table: cashing systems  - Array: math operations  - heap: task scheduling  - tree: keep the HTML document, or for AI decision  - suffix tree: for searching string in a document  - graph: for tracking friendship, or path finding  - r-tree: for finding the nearest neighbor  - vertex buffer: for sending data to GPU for rendering Over to you: Which additional data structures have we overlooked? -- We just launched the all-in-one tech interview prep platform, covering coding, system design, OOD, and machine learning. Launch sale: 50% off. Check it out: https://lnkd.in/gsxffnJE #systemdesign #coding #interviewtips  .

Essential Git Cheatsheet! 🔧 Basic Commands git init – Initialize a new Git repository. git clone <repo_url> – Clone a remote repository. git status – Check the status of your working directory. git add <file> – Stage changes for commit. git commit -m "message" – Commit staged changes with a message. git push – Push your local commits to the remote repository. git pull – Fetch and merge changes from the remote repo. git diff – Show changes in the working directory (uncommitted changes). git diff --staged – Show changes between the staging area and last commit. 🛠️ Branching & Merging git branch – List branches. git branch <branch_name> – Create a new branch. git checkout <branch_name> – Switch to another branch. git checkout -b <branch_name> – Create and switch to a new branch. git merge <branch_name> – Merge a branch into the current one. git branch -d <branch_name> – Delete a branch after merging. git branch -D <branch_name> – Forcefully delete a branch, even if it hasn’t merged. 🔄 Synchronization git fetch – Download changes from remote without merging. git rebase <branch> – Reapply commits on top of another branch to maintain linear history. git pull --rebase – Fetch and reapply your changes on top of the latest remote changes. git remote add <name> <url> – Add a new remote repository. 🎯 Advanced Git git stash – Temporarily save changes without committing. git stash pop – Reapply stashed changes. git cherry-pick <commit> – Apply a specific commit to your current branch. git log --oneline – View simplified commit history. git reflog – Show the history of your reference changes (e.g., checkout, resets). git log --graph --decorate --all – Show a visual commit history. 🚨 Undoing Changes git reset <file> – Unstage a file. git reset --soft <commit> – Reset to a commit but keep changes in the working directory. git reset --hard <commit> – Completely reset to a previous commit, discarding changes. git revert <commit> – Create a new commit that undoes a specific commit. ⚙️ Collaborating with Others git fork – Fork a repository on GitHub (via UI) to start contributing. git pull origin <branch> – Pull changes from the original remote branch. git push origin <branch> – Push your branch to the original repository for collaboration. Over to you: did we miss anything? -- We just launched the all-in-one tech interview prep platform, covering coding, system design, OOD, and machine learning. Launch sale: 50% off. Check it out: https://lnkd.in/gsxffnJE #systemdesign #coding #interviewtips  .

Netflix Tech Stack - (CI/CD Pipeline)    Planing: Netflix Engineering uses JIRA for planning and Confluence for documentation.    Coding: Java is the primary programming language for the backend service, while other languages are used for different use cases.    Build: Gradle is mainly used for building, and Gradle plugins are built to support various use cases.    Packaging: Package and dependencies are packed into an Amazon Machine Image (AMI) for release.    Testing: Testing emphasizes the production culture's focus on building chaos tools.    Deployment: Netflix uses its self-built Spinnaker for canary rollout deployment.    Monitoring: The monitoring metrics are centralized in Atlas, and Kayenta is used to detect anomalies.    Incident report: Incidents are dispatched according to priority, and PagerDuty is used for incident handling. -- We just launched the all-in-one tech interview prep platform, covering coding, system design, OOD, and machine learning. Launch sale: 50% off. Check it out: https://lnkd.in/gsxffnJE #systemdesign #coding #interviewtips  .

How does Git work?    To begin with, it's essential to identify where our code is stored. The common assumption is that there are only two locations - one on a remote server like Github and the other on our local machine. However, this isn't entirely accurate. Git maintains three local storages on our machine, which means that our code can be found in four places:    - Working directory: where we edit files  - Staging area: a temporary location where files are kept for the next commit  - Local repository: contains the code that has been committed  - Remote repository: the remote server that stores the code    Most Git commands primarily move files between these four locations.    Over to you: Do you know which storage location the "git tag" command operates on? This command can add annotations to a commit. -- We just launched the all-in-one tech interview prep platform, covering coding, system design, OOD, and machine learning. Launch sale: 50% off. Check it out: https://lnkd.in/gsxffnJE #systemdesign #coding #interviewtips  .