Shivam Vashisth

How I work

I build my own tools because the team needed answers every morning. I am not a coder, so I learned whatever the work needed.

One rule covers every tool: a number ships only when a second code path agrees with the first.

Some work stays manual on purpose. The designer finishes every thumbnail, and when a faculty member is stuck on a script, we talk it through on a call.

Where the systems get built

Claude logo

Claude Code

Every system on the Results page was built here. Each project has its own rules file, memory, skills and a retro.

The weekly tracker, the daily competitor report, the playlist rebuild and the repurposing engine.

Codex

A second coding agent I use alongside Claude Code.

Listed on my word; there is no public example to link yet.

Anthropic logo

Claude API and Gemini API

Chapters and descriptions written from transcripts, clip picks and language tags, with a fallback when a key hits its cap.

1,586 chapters written from transcripts; clips picked from 20 interview episodes.

Python logo

Python

The pipelines themselves: the pulls, the second-path verification, and the Excel and PDF files management reads.

Every weekly tracker run re-derives its numbers before it saves.

Node.js logo

Node.js

This site and the smaller build scripts, such as the cut-sheet builder.

This site is static HTML, built with Astro on Node.

Where the data comes from

YouTube logo

YouTube Data API v3

The uploads playlist as the source of truth, video details and livestream timing, with search only as a cross-check.

The weekly tracker runs on it.

YouTube logo

YouTube Studio exports

Every number I report is re-derived from a Studio export, a second way, before it ships.

The 2025 against 2026 audit behind the Results page.

vidIQ and its MCP server

Outlier ranking, keyword demand and channel comparisons, called from Claude Code through vidIQ’s MCP server.

Competitor teardowns and ranked idea lists.

yt-dlp

Metadata, comments and captions at scale when the API would cost quota, paced to stay under YouTube’s bot wall.

The 38,682-video study and the zero-cost tracker kit.

Whisper

Local transcription on my Mac. Word timings give me the timecodes for cut sheets and chapters.

Cut-sheet timecodes pinned to the exact spoken word.

FFmpeg logo

ffmpeg

Cuts, audio extraction and contact sheets when a model cannot watch the video for me.

Contact sheets read by eye when an API key ran out.

How things connect

Model Context Protocol logo

Model Context Protocol servers

Tool servers a model can call directly: vidIQ, Google Drive, Docs, Gmail and Calendar, plus one I built myself.

One MCP server built from scratch, for a charting app.

Google Docs logo

Google Docs and Sheets APIs

Script briefs land in each faculty member’s live Google Doc with real formatting; trackers export to sheets the team reads.

124 briefs in one month, none of them copy-pasted.

Google Chrome logo

Claude in Chrome

Drives admin panels that have no API, such as a store setup and a large Google Drive inventory.

It also set up this site’s hosting and domain.

Supabase logo

Supabase

Insert-only tables behind the forms on my other site, with row-level security on.

The subscribe list on shivamvashisth.com.

Vercel logo

Vercel

Every site and dashboard I run deploys from a git push.

This site, my tools site, a file vault and a private dashboard.

GitHub logo

GitHub

One private repository per project, with its rules and its history.

Every system on the Results page has its own repository.

Production and live

OBS Studio logo

OBS Studio

Simulcasting faculty livestreams to more than one channel, and fixing the stream when it misbehaves mid-class.

190 livestreams in the 250 days, up from 108.

Premiere Pro

My own edits, and the editor’s side of the pipeline from an editing cohort I am partway through.

300+ videos cut for my own channel.

Canva

Posters, overlays and share cards when the designer is busy.

Posters and countdown overlays for three mock-test events.

Shopify logo

Shopify

A storefront for the brand’s courses so they can be tagged in YouTube videos, with checkout links and policies.

Four course products set up, waiting on launch.

APIs I have worked against directly: the YouTube Data API, Google Docs and Sheets, the Claude and Gemini APIs, Supabase, the Vercel API and the Instagram Graph API.

Logos belong to their owners and are shown to say what I use, nothing more.