Kirti Purohit

AI Consulting & Software Freelancer | AI Blogs & Tips |

[Tim Denning @Tim_Denning]

Be delusional.

[Kirti Purohit @coderess_]

Make Claude speak in your way of text style & tone with this underrated skill https://t.co/mMVxY7AP2B

[Kirti Purohit @coderess_]

Reach out to Srijit here for PM mentorship!

[Srijit @SrijitSRajput]

Back to PM mentorship 🚀

After mentoring 500+ mentees globally, I paused due to workload. I’m now back to help aspiring PMs move faster — resumes, interviews, career switches & product thinking.

Trying to break into PM? Reach out👇

https://t.co/TIhcbV6KHk

#ProductManagement

[Kirti Purohit @coderess_]

Two things that make coding agents worse:

- giving them more context
- running more of them at once

Both feel like they should obviously help.

Wrote up what actually goes wrong in each case
https://t.co/GWDexbTBq3

[Kirti Purohit @coderess_]

Toy agent: generate → done

Engineering-grade agent: generate → execute → observe → test → fail → correct → verify → done

A model can say "the bug is fixed." A test suite can say "the bug is fixed."

Only one of those is checkable.

[Kirti Purohit @coderess_]

The best structural trick in coding agents: don't make one agent do everything.

The main agent holds the problem. Smaller agents go read the codebase, check security, analyse the tests — each in its own context. The main agent gets the conclusions, not the 50 files.

[Kirti Purohit @coderess_]

It's parallel research with LLM workers.

Caveat nobody posts: point three of them at the same file and you haven't built a research team, you've built a race condition.

[Kirti Purohit @coderess_]

@PratikSThombare @Flipkart @flipkartsupport True

[Kirti Purohit @coderess_]

@mohdsoftwaredev Hi Mohammad,
Happy Eid! I'm looking for Senior/SDE2 roles currently in Bangalore. Here's my resume, would loved to be vouched for, thanks!!

https://t.co/NJBTF9ajCu

[Kirti Purohit @coderess_]

@kipupwidanshika People here would pay even more if you don’t

[Kirti Purohit @coderess_]

Finally the weekend is here! 🎉

Taking a short break but still squeezing in some time to work on my iOS app.
What are your plans this weekend?

[Kirti Purohit @coderess_]

@itsAlishaPandey Congrats! Let’s connect

[Kirti Purohit @coderess_]

@N_and_ni Wow such amazing advices 🙌🏻

[Kirti Purohit @coderess_]

@yuqih Oh girl, i wish i was in SF 🤍

[Kirti Purohit @coderess_]

@yuqih I ship legal, medical field ground level projects in India, utilising AI to build software which increases transparency for people to trust and know more about systems
Let’s say, insurance and hospital documents verification and rights understanding

[Kirti Purohit @coderess_]

Banking AI chatbots come in two flavors and both are sh**:
Too tight: "I can't share that for security reasons" when you ask for your own name.
Too loose: blocks your card because a prompt said the "magic word"
Stop tuning the prompt. The LLM shouldn't have decision power over destructive actions in the first place.

The fix isn't better guardrails. It's:
→ regex routes the destructive intent
→ state machine forces confirmation
→ LLM only writes the response copy

The model should be the last thing you trust :)

[Kirti Purohit @coderess_]

A staff engineer asked me this in an interview at Global payment & credit cards org meanwhile hiring for a GenAI senior role, LOL

What’s your take on this?

[Kirti Purohit @coderess_]

How often does a senior software engineer gets asked to WRITE git commands in interviews?

Git commit, stash or rebase?

[Kirti Purohit @coderess_]

@AnshulSanghi1 Indeed, hybrid search enhances the existing system’s data. That’s a no brainer which comes prior moving to AI

[Kirti Purohit @coderess_]

RAG was supposed to kill hallucinations.
Yet in 2026, most financial AI chatbots are still confidently lying about your credit card transactions.

The “just add more context” crowd is coping hard.
Here’s what actually works in production 👇