blog .

How Infai evolved its AI agent session store from a simple JSONL log into a durable, branchable, lazy-loaded, and compactable timeline.

I swapped llama.cpp for vLLM on my local code completion setup. 49% faster generation, 36% lower latency — but 2× the VRAM. Here's what I gained, what it cost, and the gritty install details.

Using Qwen2.5-Coder-1.5B with llama.cpp and Minuet AI to get Fill-in-the-Middle code completions in Neovim — no subscription, no telemetry, running on a laptop GPU.

What building Kubmin taught me about SaaS system design, observability, deployment, queues, auth, and using AI agents without losing ownership.

How re-questioning assumptions turned a complicated MariaDB backup pipeline into a simple disk-snapshot workflow on Azure.

Skip the paid WAF. CrowdSec gives one VPS — or a whole fleet — crowd-sourced intrusion prevention with a single decision pane. Plus the multi-server trap that costs people an afternoon.

A fly.io take-home gave me an excuse to go deep on what containers actually are. What I learned about devicemapper thinpools, copy-on-write snapshots, and why a real orchestrator is just an FSM walking a state diff.

The upgrade that broke everything — then revealed how much simpler config can be.

GGUF vs safetensors, quantization, distillation, VRAM math, and Ollama vs vLLM vs llama.cpp — one post to save you the weekend I lost.

Six attempts, three days, one stubborn MX330. A post-mortem on laptop GPU passthrough in Proxmox — what failed, why it's architecturally impossible, and exactly what to buy instead.

Turn an old laptop into a full Proxmox home server — Wi-Fi NAT routing, DHCP, MTU fixes, and every gotcha you'll actually hit. A practical guide that also teaches you how AWS works.

A developer's hands-on guide to quantum computing fundamentals — from qubits and gates to building a quantum teleportation circuit, no physics degree required.

A Docker networking war story — why localhost lies inside containers, how --add-host gets you close, and how socat bridges the final gap for Playwright E2E tests.

A practical guide to Merkle hash chain audit logs — what they are, how they work, and the honest question most teams skip: do you actually need one?

A cautionary tale for DevOps engineers about the dangerous shortcut of chmod 666 on Docker sockets, and the proper GID-based approach for Docker-in-Docker setups.

Go's new runtime/secret package addresses a blind spot in secret management — what happens to sensitive data after it's loaded into memory. A deep dive into when and how to use it.

I went down a rabbit hole figuring out why my nodes had less resources than expected. Here's everything I learned about kubelet reservations, the allocatable formula, and drop-in configs.

Practical insights and best practices for managing PostgreSQL databases and implementing effective caching strategies.

A technical autopsy of Frappe's architectural patterns — ORM string-stitching, permission sprawl, memory leaks, and no-rollback migrations — backed by development experience of more than 1 year and maintainer admissions.

A developer's practical guide to instrumenting applications with OpenTelemetry, sharing key learnings on spans, context propagation, and sampling.

How to resolve network connectivity issues between Docker and virt-manager VMs

A recap of my exciting tech journey in September 2025

Configuration for making Delve work with neovim

Created a working project using the wasi's new composition feature