PinnedInDatummbyVAIBHAV GANDOTRA·Aug 11Testing Meta’s New Muse Glimmer Against Opus 5 On My Interview ScoringThe scores came out one point apart, which I think turned out to be the least useful thing I measured.
VAIBHAV GANDOTRA·Aug 22LexBench: A Benchmark for Evaluating Frontier LLMs on Indian Rental Agreement AnalysisOverview
VAIBHAV GANDOTRA·Jul 28BARS: The Rubric Trick From 1963 That Fixes Vague EvaluationsEvery hiring manager has sat through a debrief where two interviewers rate the same candidate a 4 and a 2 on “problem solving,” and neither…
InDatummbyVAIBHAV GANDOTRA·Jul 24NVIDIA’s Trick to Acing Kaggle Competitions With Agentic CodingRecently I took part in a Kaggle Playground competition, Predicting Stellar Class, and used it to test how far agentic coding tools like…
VAIBHAV GANDOTRA·Jul 22Domain Interviews, a Nasty Production Bug, and Letting AI Stop Pretendingmockintervew.com the AI mock-interview app I’ve been building for students prepping for placements got a full day of shipping today. Five…
VAIBHAV GANDOTRA·Jul 20I Took a Competitor’s Mock Interview So I Wouldn’t Have to Guess What They’re Doing WrongMost competitive analysis in this space is done from the outside — landing pages, pricing tiers, a scroll through the feature list. I…
VAIBHAV GANDOTRA·Jul 19Okay, I Think I’ve Been Talking to My Users WrongSo I read The Mom Test this week, and it kind of wrecked my afternoon in a good way.
InDatummbyVAIBHAV GANDOTRA·Jul 17The Baseline I Actually Picked for My Kaggle Pokémon Agent, and WhyBefore writing a line of my own agent for The Pokémon Company’s PTCG AI Battle Challenge, I pulled six community notebooks to see what…
VAIBHAV GANDOTRA·Jul 15Where I Actually Used AI Agents in My Interview-Prep App (and Where I Deliberately Didn’t)Every “I built an AI agent” post I read makes it sound like the whole system is agentic one big autonomous loop making decisions end to…
VAIBHAV GANDOTRA·Jul 13I Let v0 Design My App and Claude Code Build It. Here’s the Division of Labor That Actually WorksI was three chat messages into building the recording screen for a video-interview app when v0 stopped me cold: “Maximum context limit…