The final part of the divvy-forge series: wiring TrueForge deployment, discovering the real agent manifest API through trial and error, fixing coordinator prompt flow so subagents run silently, and watching the agent open a real GitHub PR with coordinator reasoning, a unified diff, and subagent findings.

How I built two MCP tool servers — divvy-reader and market-data-fetcher — that give a TrueForge agent live access to a GitHub portfolio and real stock fundamentals. Includes a walkthrough of the raw MCP protocol, BeautifulSoup scraping, and the yfinance fallback.

How I designed the coordinator and subagent prompts for divvy-forge as versioned markdown files, embedded subagent instructions inline, defined strict output contracts, and used promptfoo evals to catch three real prompt bugs before a single live agent run.

A practical walkthrough of getting TrueForge running locally, wiring up Daytona as the sandbox provider, and registering a minimal agent via the REST API — including every non-obvious error I hit and how I traced each one.

Web scrapers break silently when a site changes its DOM. Here’s how I built a pipeline that detects a broken scrape, analyses the live page, and rewrites its own parser automatically — using Bright Data Scraper Studio’s API.

How to add LiteLLM as a proxy layer on top of Ollama and Open WebUI, unifying local and cloud AI models behind a single OpenAI-compatible endpoint, managed via GitOps with ArgoCD.

How to wire the LiteLLM gateway into aider for terminal-based AI pair programming, using model aliases, a port-forward systemd service to work around SSL issues, and the current 3B model lineup running on a GTX 1650.

The second part of the GPU homelab series: bootstrapping a full GitOps platform on k3s using ArgoCD’s app-of-apps pattern, wiring up Ollama and Open WebUI, and the four gotchas that will trip you up.

A step-by-step walkthrough of turning an old GTX 1650 laptop into a GPU-enabled single-node k3s cluster, explaining the why behind each layer of the stack and the two non-obvious gotchas that will trip you up.

A practical guide for software engineers on building production-grade RAG systems using hybrid retrieval, re-ranking, and grounding techniques to reduce hallucinations and improve answer quality.