Blog

I'm building five products in public — InstantPlan, AllOS, BankGaadi, RecoverOS, and Viksepa. Here's the operating system I use to ship multiple AI products solo without drowning, and why building in public is a product strategy, not a marketing tactic.

I crossed a 3000 puzzle rating on Chess.com — the top 0.15% of players. The skills that get you there are the same ones product strategy demands: calculation under constraint, pattern recognition, and seeing several moves ahead.

The biggest mistake AI PMs make is shipping an LLM feature with no way to measure whether it's good. Here's how I build the evaluation harness before the feature, and why it's the highest-leverage thing you can own.

Standard RICE prioritization under-weights the two things that decide whether an AI product survives: habit impact and AI-readiness. Here's the weighted variant I use, with real scoring examples.

Multi-agent systems are the buzzword of the moment. Most products don't need them. Here's how I decide when multiple agents beat one good prompt, and how to scope an agent system as a PM.

I shipped Listen2RE's first version in 8 weeks instead of the 6 months a native build would have taken. Here's the scoping discipline that makes an 8-week AI MVP possible, and the trade-offs you accept to get there.

RAG was supposed to stop hallucinations. Often it doesn't. Here are the five reasons retrieval-augmented systems still make things up, and the product-level fixes for each — most of which aren't model problems at all.

AI gives you a novelty spike, then users churn. Retention comes from habit design, not model quality. Here's how I build the trigger-action-reward-investment loop into AI products, with what worked on Listen2RE.

PMs don't need to out-prompt engineers. They need enough fluency to scope what's possible, write a good spec, and judge output quality. Here's the working framework I use, including role-based prompting that simulates senior reasoning.

Most AI dashboards measure the model and miss the product. Here's how I separate AI quality metrics from product health metrics, the anti-metrics I refuse to optimize, and how to build a north-star tree that connects them.

Most articles on RAG vs fine-tuning are written by ML engineers. Here's the PM's perspective: when each approach makes sense, what the trade-offs actually cost, and how to make the call.

Writing a PRD for an AI feature is different from a standard product spec. Here's the framework I use, what sections a standard PRD misses, and a real LLM PRD template you can use.

I've had 30+ conversations with AI PM recruiters and hiring managers this year. Here's exactly what they look for — and what makes most portfolios invisible.

The exact tools, APIs, and frameworks I use as an AI PM building real LLM-powered products. Not a list of buzzwords — a working stack with specific reasons for every choice.

I'm a product manager, not an engineer. Here's how I built InstantPlan — a working AI planning tool — using LLMs, no-code tools, and a first-principles approach to shipping.

A full product case study of Listen2RE — the AI-augmented audio learning platform I designed and launched for UPSC/MPSC aspirants. Problem, architecture, metrics, and lessons.