Tech Blogs Digest 15.12 - 21.12
This week we AI-analysed 8838 posts for you, filtered out the chaff and hand-picked the wheat. Subscribe to stay up to date with future digests!
This week
🏗️ Architecture - Scaling to millions transactions daily, performance optimizations to achieve low latency, and adding authentication to a Kafka client
🧮 Data Science - AI music star analysis, factory land usage compliance checks, and detecting a web of fraud companies
💻 AI-assisted coding - The limits and strengths of AI-assisted testing
📊 Data - Repartitioning a 1.1 TB default partition
🤖 LLMs in production - Fine-tuning techniques, extracting training data from a model and defending against it, and creating a GPT-like model from scratch
🛡️ Security - The king of encryption protocols for tiny devices, removing malware from a pirate streaming app, and more
🐍 Python - A memory leak journey while upgrading to 3.10
🐘 PHP - Handling concurrent HTTP requests in PHP
🏗️ ARCHITECTURE
A behind-the-scenes look at how Zepto diagnosed real latency spikes and engineered dramatic p99 performance improvements in high-throughput services
Processing Millions of Records on IBM watsonx | 9 min read
Engineers scaled an LLM API pipeline to handle millions of calls per day, tripling throughput by shifting from threads to async I/O and smarter architecture
A developer finally implements SASL PLAIN and SCRAM authentication in the Elixir kafka_ex client, enabling secure Kafka connections many years after the original request
AWS Lambda Deep Dive: The Parts No One Talks About | 12 min read
A candid, low-level look at how AWS Lambda really works under the hood - from execution environments to VPC networking, concurrency and when it’s actually the right tool
🧮 DATA SCIENCE
I Investigated the Top 3 AI-Generated Artists Going Viral on Spotify. Here’s Who They Are Imitating | 23 min read
AI “artists” like Velvet Sundown, Breaking Rust and Aventhis are engineered to mimic familiar sounds and rack up millions of Spotify streams, revealing how algorithms reward mimicry over musical depth
Detecting Factory Utilization from Space: How We Built an AI System for Industrial Land Monitoring | 13 min read
AI and satellites now inspect thousands of industrial sites automatically, segmenting built-up areas from imagery to flag land-use compliance faster than manual surveys
A hobbyist fine-tunes Qwen3 on Reddit’s dadjokes to make it reply to any prompt with a dad joke, complete with dataset creation and evaluation
I Built an AI System That Detects Financial Fraud Hidden in the Network of Companies — Here’s How | 28 min read
AI graph analysis exposes hidden corporate fraud patterns like circular trading, shell company chains and indirect control long before they show up in transactions
💻 AI-ASSISTED CODING
AI in QA/AQA | 15 min read
AI can’t yet fully replace QA engineers, but current tools and workflows can generate, run and maintain tests with AI help - showing where automation helps most and where it still falls short
📊 DATA
A massive 1.1 TB default partition slowed every “ATTACH PARTITION” to hours; this shows how pre-copying into range partitions cut lock time to just 44 seconds and fixed performance
🤖 LLMS IN PRODUCTION
LLM Fine-Tuning: LoRA vs Full Fine-Tuning — a Comparison | 15 min read
Efficient LoRA fine-tuning slashes compute and memory needs while full fine-tuning maximizes performance - choose based on task complexity and resources
Building a 163M-Parameter GPT Model from Scratch: A Deep Dive into Transformer Architecture | 13 min read
Hands-on guide to building a 163M-parameter GPT from scratch reveals transformer internals and real-world implementation lessons
Illuminating the Dark Corners of AI - Exploiting Shadow Data in AI Models and Embeddings | 35 min read
Security-focused breakdown of a DEF CON talk reveals how attackers extract private data from AI models, embeddings and RAG systems - and how to defend against it
Inside FuzzingBrain: How an LLM‑Powered CRS Detects and Patches Vulnerabilities at Scale | 14 min read
LLM-powered FuzzingBrain autonomously hunts, reproduces, and patches real code vulnerabilities at scale using advanced LLM-driven detection and patching strategies
LLM-driven reasoning meets structured domain knowledge in a Gemini 3-powered A2A agent that infers real-world insights and enforces reliable multi-agent communication
A smart request router shifts every query to the right model for cost, speed, and quality, replacing monolithic LLM calls with efficient, dynamic pipelines
A step-by-step guide shows how to take LangChain deep agents from prototype to scalable AWS production deployment with monitoring, security, and parallelised workflows
AI’s next big hurdle isn’t bigger models - it’s how humans and systems truly understand and collaborate with each other through better interaction design
🛡️ SECURITY
IoT devices get a powerful yet tiny cryptographic engine with Ascon, offering strong authenticated encryption and hashing tailored for constrained, low-power environments
Malvertising ads trick developers into downloading a fake DockerDesktop.exe trojan that uses GPU-gated decryption and staged payloads to evade detection and install malicious code
Dissecting Android Malware: How I Neutralized a Residential Proxy Botnet inside a Fake Netflix App | 18 min read
Fake Netflix-branded Android malware secretly turns infected devices into residential proxy bots, siphoning bandwidth and hiding its payload through deep binary manipulation and persistence tactics
AETHER — AI Engine To Help Engineers Reverse | 13 min read
Aether AI augments reverse engineering with automated insight extraction, accelerating vulnerability analysis and understanding of complex binaries at scale
🐍 PYTHON
Lyft’s Python 3.10 upgrade uncovered a subtle memory leak that spiked latency, leading engineers through real-world debugging with profiling tools to find and fix the root cause
🐘 PHP
Simulating Сoncurrent Requests: How We Achieved High-Performance HTTP in PHP Without Threads | 14 min read
ManyChat engineers unlocked simulated concurrency in PHP using smart request serialization, non-blocking I/O, and curl_multi to handle high-performance HTTP without threads or extra processes
Incredible curation on the security section especially the AETHER write-up! The intersection of AI and reverse engineering feels like where the field is headed—automating the grunt work of binary analyis while letting engineers focus on the creative exploitation aspects. I remember grinding through IDA Pro sessions for hours before tools like this even existed; now ML models can highlight patterns we'd miss. Really appreciate how the Android malware piece shows real-world evasion tactics instead of just theory.