How Transformer Self-Attention Actually Works in LLMs
A step-by-step breakdown of Query, Key, and Value vector matrices, scaled dot-product attention, and why Transformers replaced old RNNs.
Welcome to Mashae TechMind. Here I publish practical engineering guides, real-world architecture breakdowns, and interactive technical quizzes. No corporate fluff—just clear code, performance benchmarks, and software engineering notes on AI, DevOps, security, and web runtimes.
Real-world engineering concepts I document across modern software stacks.
Breaking down Transformer self-attention, token embeddings, GPU matrix multiplication, and practical prompt/RAG context injection.
Comparing Linux cgroups, namespaces, and kernel sharing against hypervisor virtual machines for fast, lightweight deployments.
Enforcing strict identity controls, WebAuthn Passkeys, mTLS micro-segmentation, and least-privilege security policies.
High-dimensional vector indexing, Cosine distance metrics, HNSW graph search, and building production RAG pipelines with pgvector.
Compiling Rust, C++, and Go into Wasm binary modules to run computationally heavy workloads inside browser runtimes at native speeds.
Deep-dive engineering notes written for developers, architects, and technical learners.
A step-by-step breakdown of Query, Key, and Value vector matrices, scaled dot-product attention, and why Transformers replaced old RNNs.
Understanding host kernel sharing, Type-1 vs Type-2 hypervisors, memory overhead, and multi-stage production Dockerfiles.
Moving past firewall perimeters. How continuous identity verification, WebAuthn biometrics, and micro-segmentation stop attacks.
How vector similarity search works under the hood, comparing Cosine and Dot Product metrics, and querying PostgreSQL pgvector.
How Wasm binary bytecode bypasses JavaScript JIT overhead to execute heavy math and rendering tasks near native hardware speed.
Try out 5 quick technical questions on AI, cloud containerization, security, and web performance.
🎮 Take Interactive Quiz