Writing
Notes on things we've worked on or thought hard about. No gated downloads, no email required.
What the GameStop squeeze actually says about market structure
How a retail-driven liquidity crisis rewrote assumptions that had gone unquestioned for decades.
How to structure a multi-tranche round without losing the company
Term sheet mechanics for hardware and deep-tech founders, where the standard structures fit badly.
Why comparable-company valuation falls apart for pre-revenue hardware
The most common valuation method, and the specific situations where it produces a confidently wrong number.
We tested seven forecasting models on one stock. Only one survived.
What out-of-sample testing actually filters out, and why in-sample accuracy is close to meaningless.
A regime model, six crashes, and what it saw coming
Building a hidden Markov model over macro indicators, and honestly assessing what it caught and what it missed.
The A/B test that worked on desktop and did nothing on mobile
Why splitting your results is not optional, and the guardrail metric almost everyone forgets.
Bottom-up market sizing, and the ₹324 crore hole it found
Two ways to size the same market. When they disagree, one of them is wrong, and that gap is the interesting part.
The first three things a small business should automate
Not the exciting ones. The boring, expensive, every-single-day ones.
What we check before we ship anything an LLM writes
The four-layer safety pipeline we use, what it catches, and the failure modes it doesn't.