ThumbnailTest.
YouTube creators run real A/B tests on their thumbnails and titles, and keep the version that gets clicked.
- Senior Software Engineer @ PXCH Holding
- 2024 — 2026
- SaaS Platform
The hard problems are rarely the obvious ones.
ThumbnailTest helps YouTube creators run real A/B experiments on their thumbnails and titles. I joined as a junior working alongside an experienced lead, learned the codebase and YouTube API intricacies, then eventually became the sole developer responsible for the entire frontend and new feature development.
New users dropped off before running their first test
A/B testing felt intimidating and complex.
Designed an onboarding flow with interactive demo showing that running a test only takes 5 steps. Made the process feel approachable. Was part of the team that scaled the product from $15K to $25K MRR.
Where the real difficulty lived
YouTube API & Extension Architecture
Maintained a Chrome extension that scrapes YouTube by replicating internal API request patterns. Not a simple REST integration — required understanding YouTube's undocumented request structures and keeping up with changes.
RAG Pipeline Architecture
Built end-to-end: Hono/Bun backend receives channel ID → fetches last 20 videos via YouTube API → extracts and cleans transcripts → generates embeddings → stores in Pinecone. Includes rate limiting and resource monitoring.
Cross-Browser Extension Porting
Adapted Chrome extension to Firefox (Manifest V2/V3 differences) and Safari (Xcode packaging, native container app). Each platform had its own gotchas around permissions and API compatibility.
Ported Chrome extension to Firefox & Safari, navigating platform-specific API quirks
Each platform had its own gotchas around permissions and API compatibility.
Going from 'the new dev' to 'the person who owns this' was the real growth.
- Grew from junior to senior, taking full ownership of frontend and feature development
- Part of scaling the product from $15K MRR to $25K MRR
- Designed onboarding flow with interactive demo — improved user retention
- Reverse-engineered competitors to improve AI title generation and propose new features
- Built RAG chatbot backend: Hono/Bun service → YouTube API → transcript processing → Pinecone embeddings
- Ported Chrome extension to Firefox & Safari, navigating platform-specific API quirks
This was my first major project, and it taught me that the hard problems aren't always the obvious ones. The YouTube API scraping taught me more about reverse engineering and API patterns than any tutorial could.