Iván's Recent LinkedIn Posts

I built my first production ML model 8 years ago. Back then with TensorFlow, image classification, forecasting models, route optimization - using the RIGHT technology for each problem. Today? Everyone's trying to solve every data problem with generative AI. It's like using a hammer for every task. In my first demos with prospects, I spend half the time separating what their problems actually need: • Generative AI ✓ • Classical ML ✓ • No ML at all ✓ Here are the reality checks: 📊 Forecasting your sales? Don't use GenAI—use time series models that have worked for decades. 📈 Analyzing CSV data? GenAI understands your query, but pandas does the math (and does it better). 🖼️ Image classification? Classical ML models are faster and more accurate than VLLMs for this specific task. We're at the peak of the Gartner hype cycle. GenAI feels magical, but it's not universal. The best AI solutions combine technologies: •GenAI translates user intent •Classical algorithms process the data •Deterministic software delivers results Example: We built a system that analyzes financial data. GenAI understands "show me Q3 revenue trends by region," but our database connector pulls the data using SQL, and Python libraries do the actual calculations. GenAI then explains the results in plain English. Not all AIs are created equal. The right tool for the right job will always outperform the trendy tool for every job.

Everyone talks about AI use cases. Let me show you what actually works in practice. I use Zylon every day for two specific workflows: Hiring and sales preparation. Not because these are the only possibilities. Because these solve real, recurring problems I have. Hiring Workflow: Step 1: Create private project for each role Step 2: Upload candidate CVs and interview transcripts Step 3: Before any interview, ask: "Summarize John's background and flag any concerns" Real output I got this morning: "John has strong React experience but no TypeScript. Previous role was backend-heavy. Ask about frontend scaling challenges and willingness to learn new frameworks." Why this works: → Private candidate data stays internal → Consistent evaluation across interviews → No manual CV re-reading before calls Sales Preparation Workflow: Step 1: Create project for each major prospect Step 2: Upload call transcripts, emails, proposal docs Step 3: Before follow-up meetings: "What were their main concerns and our agreed next steps?" Real output from yesterday: "They mentioned budget concerns around Q4. You promised pricing options and technical deep-dive. They're evaluating 2 other vendors. Focus on differentiation and flexible payment terms." Why this works: → No lost context between calls → Specific preparation for each conversation → Better continuity across team members What doesn't work: Generic AI prompts: "Help me with hiring" Vague requests: "Analyze this candidate" Public tools with sensitive data What does work: Specific, repeatable processes Private data environments AI that learns your patterns over time The difference between theoretical and practical AI: Theoretical: "AI can revolutionize everything" Practical: "AI helps me prepare better for this specific meeting" Start with processes you repeat weekly. Build AI workflows around real work patterns. Measure impact by time saved, not features used. The best AI implementations solve boring, specific problems really well. What repetitive work processes could benefit from AI assistance in your role?