Evaluating LLM Coaching Quality: Our Rubric, Our Tools, Our Mistakes
Why LLM-as-judge alone is not enough for a coaching product, the rubric we use across human and automated review, and the evaluation mistakes we wish we had avoided.
— /journal
Engineering notes & field stories
Production-tested patterns, product stories from Blink AI, Nexia Academy and Wallnetic, and the trade-offs we hit shipping real software. Written by senior engineers, not marketing.
-
-
Per-User Memory in Multi-Tenant AI: Embeddings That Do Not Cross-Pollinate
How we isolate per-user memory in a multi-tenant AI product so context follows the right person across sessions — without ever leaking across tenants or users.
-
From GPT to Claude: Switching LLM Providers in Production Without User-Visible Drift
How to abstract LLM providers behind a single interface, the differences that bite at production scale, and what we learned switching Blink AI between Claude and GPT.
-
SCORM 2004 vs xAPI (cmi5): Which E-Learning Standard Should Your LMS Use in 2026?
SCORM 2004, xAPI and cmi5 compared for 2026: tracking capabilities, LMS support, mobile, and production gotchas from a senior engineering team running Nexia Academy.
-
Nexia Academy: Building a White-Label Learning Platform
The engineering story behind Nexia Academy — moving from single-tenant to multi-tenant, white-label theming, SCORM/xAPI ingestion at scale, and what we would do differently starting over.
-
SCORM Authoring Tools Compared: Articulate vs Captivate vs Rise
A practical comparison of Articulate Storyline 360, Adobe Captivate and Articulate Rise — when each is the right tool, where they hide costs, and how their SCORM/xAPI output differs.
-
AI Safety in Corporate Training: Hallucinations and Guardrails
Why hallucinations in corporate training products are worse than in consumer chatbots, and the specific guardrails we put in place across Blink AI and Nexia Academy to catch them.
-
Multi-Language SCORM Packaging: i18n in E-Learning
How to ship multi-language e-learning content cleanly — packaging strategies, manifest design, content-author workflow, and the gotchas that show up when SCORM meets i18n.
-
Building a GROW-Methodology AI Coach: Blink AI Architecture
The system architecture behind Blink AI — prompt orchestration, session state, memory isolation, evaluation pipeline and how the GROW methodology maps to a production AI service.
-
KVKK + GDPR for SaaS: A Practical Engineering Checklist
A practical engineering checklist for shipping SaaS that meets both Turkey’s KVKK and the EU’s GDPR — data flows, consent, retention, subject access requests and the differences that matter in code.