Ravi Kumar Blog Posts

Integrating AI into legacy systems is usually more about augmenting existing business processes than replacing the legacy application. The most successful approach is to start with a small, high-value use case and expose AI capabilities through APIs. Eg:  Claims adjudication in healthcare Step-by-Step approach 1) Understand the Business problem : Let's not start with AI first. Start with a business pain point.       Example: Legacy Claims System takes 10 minutes for a claims examiner to review provider notes.     Goal: Use AI to summarize notes in 30 seconds. Evaluate the Legacy System: Understand the components w.r.t Technology stack, available integration methods. Prepare a system inventory like below: 2) Build an AI Layer: Never thought of putting AI inside the legacy application directly.  For example: Benefits: 1) Decoupled architecture 2) Easier upgrades 3) Better governance 3) Prepare the Data: AI quality is always depending on the data an...

 Here’s a beginner-friendly, step-by-step walkthrough to build a simple AI application using : 1) Cursor AI - Cursor AI is a type of artificial intelligence (AI) designed to mimic human-like cursor movements in various applications, such as gaming, robotics, and user interface design. 2) Ollama server - check my blog post given below 3) Local LLM - Llama3.2  A small web app (Chatbot) where: User types a question Your app sends it to Ollama locally Local LLM responds Response appears on screen Simple Architecture for this exercise: Four clean layers, each color-coded: Blue — Browser UI (client) Purple — Node.js App (server logic) Green — Ollama Local Server (inference runtime) Yellow — Llama 3 model (the actual weights on disk) To implement this web app using Cursor AI tool& Ollama server: 1) Download and install the Cursor AI tool ==> https://cursor.com/ 2) Download and install Ollama server locally ==>  Follow my Blog post 3) Usually Olla...

The 8 steps follow a natural S-shaped flow — left to right, then right to left, then left to right again — mirroring how you'd actually build an agent: Define the purpose → Write the system prompt → Pick the LLM  → Add tools  → Set up memory →  Build orchestration  → Choose an interface → Test and improve   Here's all 8 steps using one simple example: a doctor's appointment booking agent for a clinic. Example: An AI agent that helps patients book, reschedule, or cancel appointments at a family clinic. STEP 1 Define purpose & scope Decide exactly what the agent will and won't do. Keep it focused — trying to do everything is how agents go wrong. It will do Book, reschedule & cancel appointments. Answer clinic hours & location questions. It will NOT do Give medical advice. Diagnose symptoms. Share another patient's records. STEP 2 System prompt design This tells the AI how to behave. Think of it as the clinic's instructions given to a new reception...