My guest today is a good friend and colleague - and not to forget with whom I was a co-author for a book, Paramu Kurumathur.

In this episode, Paramu discusses how his recent development work evolved from small Google Apps Script utilities copied and adapted from online examples to building AI-connected applications via APIs to tools like Gemini and ChatGPT, including enabling Q&A over his book content.

He describes surprises from “conversing” with his books—especially that LLMs retain details he has forgotten—while noting key risks such as hallucinations and the need for precise prompts. He explains learning Cursor with guidance from our colleague, Raja, discovering that it can generate code, and rapidly producing a proof of concept that maps citizens to the Government welfare schemes using PDFs, Chroma DB, sentence transformers, and queues—work that took about a week instead of months.

The conversation contrasts older development eras with today’s dependency-heavy environments, argues many SDLC intermediate steps are compressed, and highlights transferable mid-career skills in requirements and problem translation, alongside concerns about limited debugging and testing depth.

The timestamps are approximate and do not include the time for the intro. Add about 90 seconds to locate the section

00:00 Welcome and Setup

01:16 Rediscovering Coding via Apps Script

02:02 Connecting Scripts to LLM APIs

03:21 Talking to Your Own Book

05:32 Hallucinations and Prompt Control

06:50 Learning Cursor and Building a POC

09:09 Old School Dev vs Modern Tooling

12:00 AI Changes the SDLC

13:36 Testing and Trusting AI Output

15:10 Debugging Gaps and Assumptions

16:34 Setting AI Standards

17:20 Mid Career Transfer Skills

18:48 Prompting Without Hallucinations

20:21 Courses vs Learning by Doing

23:25 Overcoming First Step Fear

25:24 LLM Limits in Astronomy

28:24 Cursor for Reliable Code

29:37 Anybody Can Code Now

31:13 Next Projects and Wrap Up