Switching from React Native + Node.js (4 YOE) to Agentic AI — need roadmap advice

I have 4 years of experience as a React Native and Node.js developer. I am comfortable with REST APIs, async/await, JSON, MongoDB, authentication, and shipping production apps. I am based in India. What I have learned so far: I recently completed an AI/LLM course that covered: • Pydantic (validation, models, serialization) • LLM theory (transformers, embeddings, attention, tokenization) • OpenAI and Gemini API integration • Prompt engineering (zero-shot, few-shot, CoT, persona prompting) • Prompt formats (ChatML, Alpaca, INST) • Ollama for local LLMs • FastAPI basics • Hugging Face model deployment • Agentic AI fundamentals — built a basic CLI coding agent What I understand conceptually: I understand that an AI agent = LLM brain + tools (Python functions) + agent loop + memory (messages list). I understand RAG, vector databases, the difference between fine-tuning and RAG, and how to structure a backend with Node.js calling a Python AI agent service when needed. What I want to do: I want to transition into Agentic AI / AI Engineer roles in India. I am not looking to become an ML researcher or train models. I want to build production AI agent systems — connecting LLMs to real business data, building tools, RAG pipelines, and shipping real products. My specific questions: 1. Is my current foundation strong enough to start building real agent projects or do I have gaps I am missing? 2. What should my learning roadmap look like for the next 3–6 months given my background? 3. Which frameworks should I prioritise — raw OpenAI API first, then LangChain/LangGraph, or jump straight to frameworks? 4. What kind of projects should I build for a strong portfolio targeting ₹20–35 LPA roles in India? 5. Any specific subreddits, communities, or resources beyond YouTube that helped you in this transition? My planned first 3 projects: • Simple agent with web search + calculator tool (no DB) • Agent connected to MongoDB with RAG • Full FastAPI backend wrapping the agent with a React fr