DEV Community

# llm

Posts

đź‘‹ Sign in for the ability to sort posts by relevant, latest, or top.
LocalForge: I built a self-hosted LLM control plane with intelligent routing and LoRA finetuning

LocalForge: I built a self-hosted LLM control plane with intelligent routing and LoRA finetuning

Comments
2 min read
MCP Isn't a Model Feature. It's a Power Outlet for Your Tools.

MCP Isn't a Model Feature. It's a Power Outlet for Your Tools.

9
Comments 8
4 min read
The Actual Cost of Self-Hosting Your LLM (Nobody Does This Math First)

The Actual Cost of Self-Hosting Your LLM (Nobody Does This Math First)

Comments
4 min read
I built an LLM-powered compliance scanner that points at the actual line of code

I built an LLM-powered compliance scanner that points at the actual line of code

Comments 2
3 min read
A Minimal ~9M Parameter Transformer LLM Trained from Scratch

A Minimal ~9M Parameter Transformer LLM Trained from Scratch

Comments
2 min read
Building a fast LLM gateway in Go: Lua + pgvector

Building a fast LLM gateway in Go: Lua + pgvector

1
Comments 1
9 min read
LLM Observability tool

LLM Observability tool

Comments
1 min read
Your LLM Is Wrong. Your Codebase Is Why.

Your LLM Is Wrong. Your Codebase Is Why.

1
Comments 8
5 min read
Two Knowledge Hierarchies: Structuring Context for AI Agents and LLMs

Two Knowledge Hierarchies: Structuring Context for AI Agents and LLMs

Comments 2
4 min read
AI Duel on Building Retro RPG Quest Journal

AI Duel on Building Retro RPG Quest Journal

Comments
3 min read
Qwen3.6-Plus Benchmark: It Is Trying to Finish the Job, Not Just Win Chat Scores

Qwen3.6-Plus Benchmark: It Is Trying to Finish the Job, Not Just Win Chat Scores

1
Comments
5 min read
Context Compression and Persistent Memory Design for Terminal AI Assistants

Context Compression and Persistent Memory Design for Terminal AI Assistants

1
Comments
7 min read
qwen3.6-27b scores 77.2% on SWE-bench. the dense model is winning against MoE.

qwen3.6-27b scores 77.2% on SWE-bench. the dense model is winning against MoE.

Comments
3 min read
I built a version manager for llama.cpp using nothing but vibe coding.

An nvm-style CLI built without knowing Go

I built a version manager for llama.cpp using nothing but vibe coding.

10
Comments 8
3 min read
I blind A/B tested 40 Claude prompt codes. Only 7 actually shift reasoning.

I blind A/B tested 40 Claude prompt codes. Only 7 actually shift reasoning.

Comments
1 min read
đź‘‹ Sign in for the ability to sort posts by relevant, latest, or top.