# LLMSlim engineering notes

> Technical writing on context compression, context engineering and evaluation.

- [LLMSlim vs LLMLingua-2 vs LongLLMLingua: A Reproducible Prompt Compression Benchmark](https://www.llmslim.app/articles/llmslim-vs-llmlingua-2-vs-longllmlingua.md): One reader, one token budget, 300 paired cases on NaturalQuestions, and every script and per-case result published
- [Graph Centrality & TF-IDF Vectorization for In-Context Redundancy Reduction](https://www.llmslim.app/articles/how-prompt-compression-works.md): Mathematical Derivation of LexRank Stationary Distributions and Priority Tier Filtering
- [Quadratic Attention Scaling O(N^2) & In-Context Token Reduction Economics](https://www.llmslim.app/articles/attention-economics-and-complexity.md): Deriving Computation Savings in Transformer Self-Attention Layers
- [Mitigating U-Shaped Attention Recall Decay in Long Context Prompts](https://www.llmslim.app/articles/lost-in-the-middle-mitigation.md): Structuring Information Density to Overcome Position-Dependent Attention Loss
- [Deterministic Safety Shields: Priority Tier Protection Mechanisms](https://www.llmslim.app/articles/priority-tier-protection.md): Formal Syntactic Safeguards for Directives, Code Blocks, and Structured Data
- [AST Syntax-Aware Normalization for JSON, XML, and YAML Prompts](https://www.llmslim.app/articles/structured-output-normalization.md): Compressing Structural Data Payloads Without Invocation Failures
- [Architectural Comparison: Offline Graph Methods vs. Neural Perplexity Pruning](https://www.llmslim.app/articles/offline-graph-vs-neural-compression.md): Evaluating Throughput, Hardware Footprint, and Latency Metrics for Edge & Cloud Gateways
- [Reverse Proxy Gateway Integration: Context Compression in Production Python Gateways](https://www.llmslim.app/articles/enterprise-gateway-integration.md): Deploying Transparent Pre-Dispatch Context Optimization in Enterprise Services

Canonical page: https://www.llmslim.app/articles