Writing · 9
Build logs, teardowns, essays
Long-form writing about building AI products — from someone doing it, in India. Shorter notes live in TIL.
- Let the field overrule the roadmapWhen observed user behaviour should change what you build, and how to make field research cheap enough to run every cycle, from Lokal to a newsroom.
- Killing a content cluster on evidenceA question-page cluster built on a sound theory, then deindexed by Google. When to stop a reasonable bet, what counts as a verdict, and what to keep.
- Running a one-person product portfolioHow I split one person's evenings across several live products: what gets built, what gets automated, what gets killed, and the metric that decides.
- Shipping native apps as a PM, with AI writing the codeWhen a PM should build the product themselves, and what changes about scope, quality and ownership once AI writes the code and the release is yours.
- Hybrid LLM Evaluation: Let Code Count, Let Models JudgeWhich parts of 'good' to give code and which to an LLM judge: per-dimension weights, hard caps, and publish bands as product decisions.
- Language Is Product, Not LocalizationBuilding for India's languages at ShareChat, Lokal, and a multilingual newsroom taught me that language changes the product, not just its strings.
- LLM Evals for Product Managers: A Field GuideA practical guide to LLM evaluation for PMs — decision-first design, golden sets, judge calibration, hybrid scoring, and regression protection.
- The Day My LLM Judge Drifted — and How I Caught ItMy LLM judge quietly inflated scores and re-ranked my configs. The before/after data, the fix, and what a PM should require before trusting a judge.
- When Constrained Generation Beats Open GenerationLessons from shipping AI in a multilingual newsroom and running an eval-gated content pipeline: constrain the failure surface, not the model's talent.