Writing · Log

Notes from the margins
of the editor.

Short essays on the software I build, the decisions that didn't make the design doc, and the unglamorous work that keeps production quiet. Updated when I have something to say.

VIM: quick look up

Essential commands, motions, and muscle-memory patterns for daily VIM usage.

~ 3 min read→

Think Before You `install`: How a Git Clone Can Compromise Your System

How a malicious package or postinstall hook can silently compromise your system.

~ 8 min read→

Fine-Tuning Llama 3.2 1B: GPU vs CPU Deployment on Cloud Run

Comparing GPU and CPU inference paths on Cloud Run for a fine-tuned language model.

llmfine-tuningvllm

~ 10 min read→

Building a Minimalist Portfolio with SolidJS

From blank page to deployed portfolio — signals, file-based routing, and zero-runtime CSS.

solidjsportfolioweb-development

~ 7 min read→

Real-Time Data Pipelines: Go, WebSockets, and Apache Druid in Production

Moving financial tick data with Go and WebSockets into Apache Druid — lessons from production.

golangwebsocketprotobuf

~ 12 min read→

Complete Guide: Monitor FastAPI Apps with New Relic APM

Setting up APM, distributed traces, and custom dashboards for a FastAPI application.

newrelicmonitoringoptimization

~ 6 min read→

Production Observability: Deploy Grafana Stack with OpenTelemetry

OpenTelemetry collector, Loki, Tempo, and Prometheus — wired together in one compose file.

observabilityoptimization

~ 8 min read→

Why Cloudflared Beats Ngrok: Zero-Config Tunnels for Local Development

Persistent, authenticated tunnels with zero port-forwarding configuration needed.

cloudflaretunnelsdevelopment

~ 5 min read→

FastAPI Lifespan Events: Managing Resources and Background Tasks

Managing startup, shutdown, and long-lived resources without global state.

fastapilifespanoptimization

~ 6 min read→

Indexify in Action: Build Multi-Stage ML Pipelines at Scale

Building multi-stage ML document pipelines with built-in parallelism and retries.

indexifyworkflowapi

~ 8 min read→

Smart Routing for RAG: Direct Queries to the Right Model Every Time

Classifying queries at the router to choose between dense retrieval and direct generation.

llmembeddingvector

~ 7 min read→

pgvector: Scale Semantic Search in PostgreSQL Without New Infrastructure

Adding semantic search to an existing PostgreSQL database — no new infra required.

llmembeddingvector

~ 6 min read→

BAML: Type-Safe LLM Outputs with Compile-Time Validation

Compile-time validation for LLM outputs using a typed schema DSL.

llmstructured dataunstructured data

~ 5 min read→

Supabase Best Practices: Row-Level Security and Edge Functions Done Right

Row-level security patterns and edge function pitfalls worth knowing before you ship.

~ 7 min read→

Choosing the Right LLM Evaluation Framework: A Practical Comparison

LangSmith, RAGAS, and DeepEval — what each does well and when to reach for it.

llmevaluationframeworks

~ 8 min read→