Recently, I completed a learning playlist on MCP (Model Context Protocol) by Nitish Sir from CampusX 🙌 and created these short PPT notes to understand it better. In this, I’ve explained in a simple way: 🔹 What MCP is & why it is needed 🔹 Problems before MCP and how MCP solves them 🔹 Basic MCP architecture (Model, Client, Server) 🔹 How communication happens (Data Layer & Transport Layer) 🙏 Big thanks to Nitish Sir (CampusX) for making complex AI system concepts easy to understand. 📌 I’ve kept the implementation part for later — I’ll cover it after completing a hands-on MCP project and share that soon. Learning step by step & sharing the journey 🚀 #MCP #AI #LearningInPublic #CampusX #BeginnerFriendly #AIJourney

Today was all about making my AI agents feel faster, smarter, and more reliable.I focused on two powerful concepts: Streaming and SQLite Checkpointing — both essential for real production-grade systems. 1️⃣ Streaming — Making AI Responses Feel Instant I learned: - What streaming is: sending output token-by-token instead of waiting for the whole response - Why it’s important: gives users instant feedback and a smoother experience - Where it’s used: chats, multi-agent reasoning, long answers, code generation - Benefits: ✔ Faster response time ✔ Better UX ✔ Great for human-in-the-loop ✔ Reduces “waiting silence” in long tasks In simple words:Without streaming → UI feels slowWith streaming → AI feels alive And I successfully implemented streaming in LangGraph today. 2️⃣ SQLite Checkpointer — Long-Term Memory for Agents After streaming, I learned how to give my agent long-term memory using the SQLite checkpointer. Why it matters Agents need to remember: - past messages - previous actions - tool results - workflow state The SQLite checkpointer helps in: - Storing conversations - Saving workflow snapshots - Restoring previous states after a crash - Supporting multi-step reasoning across sessions This is the base of fault-tolerant, persistent AI agents. Why These Two Features Are Powerful Together ✔ Streaming → instant, smooth responses✔ Checkpointing → memory + reliability Together they make an AI agent feel: - Responsive - Context-aware - Continuous - Stable across time

“And Allah will provide for him from where he does not expect.” — Surah At-Talaq (65:2)

Today was a big leap forward. I moved beyond basic agents and learned how to build intelligent, reliable, and stateful AI systems using LangGraph. Instead of simple one-direction chains, I explored how real agents make decisions, loop through tasks, and recover from failures — just like humans. 🔀 1️⃣ Conditional Workflows — Agents That Choose Their Own Path I learned how agents can: - Make decisions based on input - Select different tool paths - Adapt workflows dynamically This is the foundation of goal-driven AI agents. 🔁 2️⃣ Iterative Workflows — Agents That Think in Loops I practiced creating loops where the agent: - Plans - Executes - Evaluates - Repeats until the job is done Perfect for planning, refinement, and multi-step reasoning. 📦 3️⃣ Persistence — Saving State Like a Real System I understood why persistence is crucial: - Saves agent state - Restores previous steps - Continues after interruptions - Enables long-running workflows This is how agents become fault-tolerant. 🛡 4️⃣ Fault Tolerance & Time Travel LangGraph allows agents to: - Recover from errors - Replay previous checkpoints - “Time travel” back to a known safe state - Avoid losing progress This is a game-changing feature for production-grade AI systems. 🧍‍♂️ 5️⃣ Human in the Loop Also learned how to pause the workflow and let a human: - Approve - Modify - Inject instructions Perfect for safety-critical applications . 🧠 6️⃣ Short-Term Memory I practiced adding temporary memory so the agent can: - Hold context during execution - Pass data between steps - Clean it once the job is done This keeps the workflow efficient and context-aware. checkout my notebook for practical working overview https://github.com/umiii-786/AI-Agent-Learning-Notebooks-/blob/main/day_3_of_AI_agent_learning_condition_Flow.ipynb https://github.com/umiii-786/AI-Agent-Learning-Notebooks-/blob/main/day_3_of_AI_agent_learning_persistence.ipynb

Understanding LLM Workflows, LangGraph Execution Model & Hands-on Practice Day 4 of my AI Agent Development journey took me deeper into how LLM-based systems are structured — and how real AI workflows are designed using LangGraph. Along with the theory, I also created a hands-on notebook, where I built simple conditional, sequential, and parallel workflows to strengthen my understanding. 🔹 LLM Workflow Types I Explored Today 1️⃣ Prompt Chaining — Sequential prompting where each output feeds the next step. 2️⃣ Routing Chaining — Choosing which LLM or chain solves which problem. 3️⃣ Parallelization Chaining — Breaking a task into sub-tasks, running them together, then combining results. 4️⃣ Orchestration Chaining — An LLM deciding which sub-model is best for each step. 5️⃣ Evaluation Chaining — One model answers, another evaluates; if not satisfied → regenerate. 🔹 Core Graph Concepts in LangGraph 🔹 Nodes — Define what work is performed. 🔹 Edges — Define how the workflow transitions. 🔹 State — A shared memory (typed dict / Pydantic model) flowing through the graph. 🔹 Reducers — Define how updates to the shared state are merged. 🔹 LangGraph Execution Model Today I learned how LangGraph manages: - Input handling - Node execution - State updates - Conditional routing - Parallel flows - Cyclic workflows And I practiced implementing all of this by writing a custom notebook that includes:✔ Simple conditional workflow✔ Sequential workflow✔ Parallel workflow This practical coding session helped me understand how real AI agents operate under the hood. 👉 Here is the notebook — check out the code and feel free to share your feedback: https://github.com/umiii-786/AI-Agent-Learning-Notebooks-/blob/main/day2_of_AI_agent_making_simple_conditional_workflow.ipynb