---
title: "Rijul Sarin (@rijulsarin) | Hashnode"
description: "Shipping AI products. Working smarter with AI. Sharing what I learn. Read the latest articles by Rijul Sarin on Hashnode."
url: "https://hashnode.com/@rijulsarin"
---

@rijulsarin

# Rijul Sarin

@rijulsarin · Seattle, USA · Joined April 2026

Shipping AI products. Working smarter with AI. Sharing what I learn.

### [How Models Learn to Behave: SFT, RLHF, and What Alignment Means for Product](https://appliedailab.hashnode.dev/sft-vs-rlhf-what-alignment-means-for-product?utm_source=hashnode&utm_medium=feed)

The model was accurate. Users called it "robotic" and "like talking to a FAQ page." The team had solved accuracy. Nobody had solved alignment.

12 min read · [appliedailab.hashnode.dev](https://appliedailab.hashnode.dev/sft-vs-rlhf-what-alignment-means-for-product?utm_source=hashnode&utm_medium=feed)

[Jun 21](https://hashnode.com/posts/sft-vs-rlhf-what-alignment-means-for-product/6a38335bc3e6bcdf076746c2)

### [How to Think Like a Senior AI PM: A Field Guide to the Interview](https://appliedailab.hashnode.dev/how-to-think-like-a-senior-ai-pm?utm_source=hashnode&utm_medium=feed)

It's not when the candidate starts talking. It's the two seconds before, when they decide what to do with the question. That's what interviewers see.

15 min read · [appliedailab.hashnode.dev](https://appliedailab.hashnode.dev/how-to-think-like-a-senior-ai-pm?utm_source=hashnode&utm_medium=feed)

[Jun 21](https://hashnode.com/posts/how-to-think-like-a-senior-ai-pm/6a3832695893629eaad9136f)

### [How to Build an AI Product That's Actually Defensible](https://appliedailab.hashnode.dev/ai-product-defensibility-real-moats?utm_source=hashnode&utm_medium=feed)

Differentiation is what makes your product better today. Defensibility is what makes it hard to replace tomorrow. They're not the same thing.

13 min read · [appliedailab.hashnode.dev](https://appliedailab.hashnode.dev/ai-product-defensibility-real-moats?utm_source=hashnode&utm_medium=feed)

[Jun 21](https://hashnode.com/posts/ai-product-defensibility-real-moats/6a383151475f137d674f7daa)

### [AI Integration for Existing Products: The Questions Most Companies Get Wrong](https://appliedailab.hashnode.dev/integrating-ai-into-existing-products?utm_source=hashnode&utm_medium=feed)

Duolingo used AI to cut costs. The result: reputational damage and 300,000 lost followers. The technology wasn't the problem. The framing was.

16 min read · [appliedailab.hashnode.dev](https://appliedailab.hashnode.dev/integrating-ai-into-existing-products?utm_source=hashnode&utm_medium=feed)

[Jun 21](https://hashnode.com/posts/integrating-ai-into-existing-products/6a382f5c394d72bd17052a82)

### [Designing AI Agents: How Much Autonomy Is Actually Safe to Ship?](https://appliedailab.hashnode.dev/designing-ai-agents-how-much-autonomy?utm_source=hashnode&utm_medium=feed)

A 10-step agent at 95% per-step reliability succeeds 60% of the time. The reliability math every PM should run before scoping an agent.

16 min read · [appliedailab.hashnode.dev](https://appliedailab.hashnode.dev/designing-ai-agents-how-much-autonomy?utm_source=hashnode&utm_medium=feed)

[Jun 21](https://hashnode.com/posts/designing-ai-agents-how-much-autonomy/6a382df06b0caf1c8b4a1550)
