@ericsiwakoti

From Model to Production: Auto-Subtitles for Vimeo & Stripe Automation

3d ago · 3 min read · Engineering post-mortems from teams shipping AI features at scale reveal a single recurring truth: success depends less on model quality and more on designing systems that anticipate and contain failu

Claude’s Inner Workings and the Mechanics of LLM Reasoning

Apr 26 · 3 min read · Interpretability work from Anthropic exposes a recurring gap; LLM explanations often do not reflect internal computations. That gap matters when you use chain-of-thought outputs for debugging, verific

Overcoming Hard-Question Failures in RAG Using Agentic Design Patterns

Apr 13 · 4 min read · I reviewed post-mortems from teams deploying Retrieval-Augmented Generation (RAG) in production and found a consistent failure mode: RAG handles simple questions well but breaks down on ambiguous quer

Beyond Prompts: Engineering Context for Reliable AI Agents

Mar 29 · 4 min read · I've been studying engineering post-mortems from teams shipping AI agents. A pattern keeps surfacing: agents work perfectly in demos but fail silently in production. The culprit isn't the model. It is

Complete Guide to Secure State Management in MERN Applications

Feb 22 · 5 min read · You build a login form. It works locally. Then you deploy. Users refresh the page and lose their session state. The UI flickers. You panic and move the JWT (security token) into localStorage. Now you