Tech Blogs Digest 03.11 - 09.11
This week we AI-analysed 8610 posts for you, filtered out the chaff and hand-picked the wheat. Subscribe to stay up to date with future digests!
This week
📊 Data - Historical and real-time data pipelines
🏗️ Architecture - Scaling database in Uber, regionalization in AWS, and fair rate limiter
🤖 LLMs in production - Video search system, production-grade patterns for AI agents, and DAG engine for AI workflows
🧠 ML - end-to-end ML system insights, and probabilistic forecasting
💻 AI-assisted coding - Metaflow from Netflix for AI and ML development, AI for mobile testing, and teaching Claude you patterns
⚙️ DevOps - Node auto-provisioner for Azure AKS
📊 DATA
Modern streaming tools meet real-time analytics: see how the team at Fresha orchestrates historical and streaming data flows using Flink, Airflow and StarRocks
🏗️ ARCHITECTURE
How Uber Scaled from 40M to 150M Database Reads/Second — Without Adding More Databases | 16 min read
From 40 M to 150 M reads/second: how Uber slashed costs and boosted performance by rethinking caching, not adding databases
When global teams need access to region-specific services, a central hub can scale seamlessly - learn how enterprise networks evolve into satellite region architectures
OpenChoreo: The Secure-by-Default Internal Developer Platform Based on Cells and Planes | 20 min read
Learn how the open-source platform OpenChoreo uses a ‘cells and planes’ architecture to give platform engineers built-in security, observability and developer abstractions
Why a Database View is the better ML-Model Endpoint | 9 min read
The article describes how turning your ML model into a database view could simplify deployment, cut latency, and make predictions instantly accessible at scale
Designing Fair Token Bucket Policies for Real-Time Apps | 21 min read
Learn how tuning burst capacity and refill rates with the token-bucket algorithm keeps real-time apps smooth for users and resilient against abuse
🤖 LLMS IN PRODUCTION
Building a Multi-Vector Semantic Video Search System with Amazon Nova Multimodal Embeddings Model | 22 min read
See how a unified multimodal model turns raw video, audio and text into searchable moments - powering semantic video search at segment-level precision
Get a peek into how production-grade AI agents actually succeed - through clear goals, context isolation, intelligent routing and validated workflows
A DAG-Based Approach to LLM Workflow Orchestration | 6 min read
Learn how rewriting your LLM orchestration with a DAG-centric engine cut hundreds of lines of messy code and stopped the 3 a.m. firefights
An AI agent taps into an optimisation engine to redesign supply chains live - balancing cost, capacity and sustainability in one conversational workflow
See how deploying a remote MCP server - complete with OAuth 2.1, Dynamic Client Registration and HTTP handshakes, unlocks seamless custom-agent integration with Claude.ai and mobile endpoints
All You Need to Know About Chunking in Agentic RAG | 10 min read
Learn how advanced chunking strategies: from semantic splits to hierarchical parent-child designs, can slash token usage and dramatically boost retrieval accuracy in agentic RAG systems
Explore how adopting agent-specific identity flows ensures credentials and secrets remain isolated, secure-by-design and resistant to misuse across your agentic applications
🧠 ML
Hands-on insights from building a real-world eye-tracking model: expect messy data, lots of iteration and one big lesson - deployment matters
Master advanced forecasting techniques: from dense quantile grids to deep learning and conformal prediction - to model full predictive distributions and deploy reliable uncertainty-aware systems
💻 AI-ASSISTED CODING
The article reveals how Netflix’s open-source framework Metaflow (with its new “spin” command) lets ML engineers iterate like in notebooks, yet ship production-ready workflows at scale
Scaling Mobile UI Testing with AI | 11 min read
By looping AI into the mobile-UI testing loop, the team transformed 4,869 tests into 10,400 in under a year - while keeping stability above 96%
From Asking Claude to Code to Teaching Claude Our Patterns: Building Modular AI Skills | 14 min read
A hands-on journey reveals that getting the most from Claude means teaching it your patterns - then watching it code alongside you
⚙️ DEVOPS
When pods go unscheduled, Karpenter-powered auto-provisioning spins up the right VM in Azure AKS - no more pre-baking heaps of node pools