04 — Transmissions
What we learn engineering agents, retrieval, and orchestration — written down plainly, for people building the same things.
KritiFin turns a financial adviser's own documents into ranked priorities, citation-backed answers, and meeting briefs. The design: proactive by default, grounded by architecture.
kritifin · financial advisers · grounded ai · rag
n8n is excellent at deterministic workflows and quietly terrible at ambiguity. Here's where the ceiling is, and when an AI agent loop actually earns its place.
n8n · automation · agents
Most RAG failures are retrieval failures wearing a generation costume. Chunking, hybrid retrieval, grounding, and the evals that catch problems before users do.
rag · retrieval · llm
RefineIQ rewrites rough AI prompts into clear ones with a language model that runs entirely in your browser. The architecture, the trade-offs, and why zero backend is the feature.
refineiq · on-device ai · privacy
Multi-agent systems fail in predictable ways. The patterns that keep them deterministic where it counts — role separation, verification gates, and budgets on everything.
agents · orchestration · llm