Llm | martinuke0's Blog

Diagram of a pi-agent control loop with memory, tools, and a verifier.

Building a pi-agent for Long-Horizon Tasks: Patterns That Actually Work

Pi-agents for long-horizon tasks need more than a clever prompt. This post covers the architecture, memory patterns, and failure modes that decide whether the agent finishes the job or burns budget on a loop.

A stylized illustration of transformer blocks and a KV cache heatmap.

Build Your Own Inference Engine: From Scratch

Demystify how LLM inference works by building a tiny inference engine from scratch. Covers tokenization, the transformer forward pass, KV cache, and decoding.

A microcontroller board beside a tiny neural network diagram.

Optimizing Small Language Models for Local Edge Inference: Techniques, Constraints, and Production Deployment Patterns

Learn practical techniques to squeeze LLMs onto edge hardware, manage resource limits, and apply proven deployment patterns.

Diagram of a multimodal RAG pipeline linking image encoder, vector store, and LLM.

Architecting Multimodal RAG Pipelines: Integrating Vision-Language Models for Production-Ready Applications

A deep dive into building production‑grade multimodal RAG systems, covering architecture, data flow, scaling, and monitoring with real‑world examples.

Illustration of a Rust crate connecting to several LLM provider APIs.

Implementing Liter-LLM: Architecting Rust-Powered Polyglot Bindings for Multi-Provider Inference and Production-Ready Pipelines

A step‑by‑step guide to designing a Rust inference engine, exposing it to multiple languages, and wiring it into a fault‑tolerant, observable production workflow.