@helkyn-coello

Helkyn Coello

@helkyn-coelloMunich, GermanyJoined June 2026

AI Lead | Embedding AI into enterprise software and internal workflows | Production AI with results

About

I have spent twenty-five years inside one company watching technology change in cycles — from desktop to client-server, to fully web-based, and now to AI-native. What that teaches you, mostly, is humility. Every wave looked transformational. Most were. But the ones that delivered lasting value were rarely the ones that arrived loudest. They were the ones that earned their place inside systems that already worked. This shapes how I think about AI today. Not as a replacement for what works, but as a layer that earns its place inside systems that already serve real customers. I lead AI company-wide adoption across products, operations and teams. My approach favors production over pilots, integrating over rebuilding from scratch, and delivery speed over endless planning cycles. Twenty-five years in, I still find this work interesting — and that surprises me as much as anyone.

Available for

AI Consulting services including: AI Audit, KB Building

Helkyn Coello's blogs

Unlocked Consultingunlockedconsulting.hashnode.dev18 posts

Comments

Farid, honestly you don't need all six of those disciplines to start. For what you're building, it mostly comes down to one thing: knowledge graphs plus RAG. That's the combination that lets an AI find the right piece of your logic under the right market conditions, which is exactly what you described wanting. The single most useful resource for your goal is DeepLearning.AI and Neo4j's short course "Knowledge Graphs for RAG" (on Coursera). It's practical and teaches precisely the retrieval problem you named: storing your ICT concepts with their relationships, not just as loose text, so the system pulls the correct model when the conditions match. Class Central also lists several free RAG courses (LangChain, LlamaIndex, DeepLearning.AI) if you want more. One honest note. The hardest part won't be storing or retrieving the knowledge. It will be the rules that connect a chart observation to a specific trade model: "when I see X on the chart, apply model Y." ChatGPT's "execution-grade" output tends to make that step sound more solved than it is. Get that mapping right and the rest follows. Happy to point you further if you get stuck there, it's the kind of thing I work on.