Vibecoding

I went to Sundance with a group of people sharing housing and meals, and built a command center app listing the schedule of movies, talks, and panels, a directory of attendees with profile pics and bios, plus housing logistics of who was staying in which house and when each person was arriving and leaving.

When I did a three-month volunteering trip traveling to film festivals in Edinburgh, Switzerland, Milan, Toronto, and South Korea to take red carpet photos for Wikipedia, I built transitpaths, a travel tracker so friends could follow along on my adventures and see where I was in the world: a timeline that knows where I am based on today's date, with transit paths between cities and a mini calendar.

I'm interviewing my dad about his stories from growing up in India in My Father's Stories. I built an audio transcription pipeline with diarization and known-speaker references, so the transcript actually says “Dad” and “Jay” instead of “Speaker 1” and “Speaker 2”, plus provider routing across multiple transcription services, chunking controls for long recordings, and an LLM-powered cleanup pass that fixes punctuation, removes filler words, corrects mishearings, and standardizes proper nouns based on a lexicon of known entities like Bombay, monsoon, kulfi, babuji, Air India, and Kashmir Princess.

I built Protagonist, a fitness dashboard that parses Apple Health XML to track my weight, workouts, and eating habits. It externalizes the part of my brain that wants to lie to itself, surfacing patterns I'd otherwise rationalize away, like weeks where I told myself I was eating well but actually wasn't, or stretches where I skipped more workouts than I remembered.

It turns abstract goals like “get in shape” into a daily feedback loop by showing me, every morning, exactly what I did yesterday and how it compares to my trend line, so the next workout or healthy meal feels like an obvious next move rather than a fresh act of willpower. I built a separate weight tracker on top of that, with a dashboard and a PDF report generator.

To stay on top of AI news, especially AI adoption, creativity, writing, and vibecoding, I built AIDB, a searchable archive of more than 1,200 AI Daily Brief transcripts, and Dario Transcripts, a separate searchable archive of Dario Amodei interviews on YouTube.

For more than ten years, I've done 100% of my writing in Emacs, and I've used Claude Code to build all kinds of Rube Goldberg architecture to support my thinking, writing, research, and personal knowledge systems, including an org-mode exporter that outputs Emacs org-mode to Markdown that Astro can render.

I built a writing streak visualizer, multiple writing tracker variants, and a writing-momentum app: tools that make the invisible parts of writing, consistency, drift, momentum, avoidance, visible enough to act on.

I built Prolific, a writing momentum system with command-line check-ins, publishing workflows, and native macOS Swift popups that ask whether I'm actually doing creative work or just cleverly avoiding it. Every 10 to 20 minutes, a floating window appears on my screen with single-keypress interaction, so it takes about one second to respond. It also runs morning check-ins, pomodoro work sprints during work sessions, and end-of-session reflections to close the day.

For rapid idea capture, I made h3-workflowy, a terminal-based outliner inspired by Workflowy: full-screen, auto-saving, and keyboard-driven, with undo/redo, drag reordering, and clipboard-friendly exports, so I can capture thoughts fast and shape them later.

I built a Readwise reading page on Socratic that automatically pulls in the AI articles I'm currently reading and annotating in Readwise, so the site shows whatever I'm reading right now instead of a static “recommended reading” list that I update once a year. The page rebuilds whenever I save a new article, with duplicates removed. It's filtered to AI-related reading specifically, not my entire saved-articles firehose, and visitors can narrow further by topic like agents, reasoning, and coding, or by publication like The New York Times, Wired, and The Atlantic.

I also built Research & Studies pages on Socratic: a browsable research library of academic studies on AI, education, creativity, and cognition, things like neuroscience studies on how ChatGPT affects essay writing, surveys on AI's impact on critical thinking, and Google's chain-of-thought paper. Each entry has a summary covering why the study matters, the research design, key findings, and limitations, plus evidence-quality badges like meta-analysis, RCT, observational, and preprint.

And I built Open Questions pages on Socratic, editorial pages built around Richard Feynman's concept of “open questions”, taking the AI articles I save to Readwise and reorganizing the reading corpus around the unresolved debates in the field, so each page pairs a live question with the strongest pieces arguing each side.