I Fine-Tuned a 9B Model Into a True Data Analyst Agent (Qwen3.5-9B + LoRA) — Base Model Failed Completely, My Version Completes 89% of Tasks Autonomously

Most local AI models claiming to be "agentic" are basically just fancy tool-callers. Give them "Analyze this dataset," and they'll do one step, stop, and wait for you to tell them what to do next. That's not autonomy—that's just following orders. So I decided to test something: can you actually make a small <10B model genuinely autonomous by training it properly, instead of building some massive external framework around it? Turns out, yes. I took Qwen3.5-9B and fine-tuned it (LoRA) on agentic workflows. The base model? Completely failed at independent analysis. My version? Handles 89% of real data analysis tasks start-to-finish without needing you to babysit it. This is the difference between a model that *looks* smart and one that actually *thinks* through problems. If you've been frustrated with local models hitting a wall, this might be exactly what you've been waiting for.