Featuring former president Barack Obama
Featuring Stewie Griffin from Family Guy

ReelJobs

Inspiration

Job hunting sucks. You scroll through endless walls of text, copy-paste the same resume into a dozen forms, and still hear nothing back. Meanwhile, we spend hours on TikTok and Instagram Reels consuming content effortlessly.

We asked ourselves: what if finding your next job felt as easy as scrolling through Reels?

Gen Z and millennials already consume information through short-form video. Why should job discovery be stuck in the 2010s? ReelJobs brings job listings into the format we actually enjoy consuming.

What it does

ReelJobs transforms job listings into engaging short-form vertical videos. Instead of reading through job descriptions, you swipe through 15-30 second video summaries that capture the essence of each role.

Key features:

Swipe-to-discover feed - TikTok-style vertical video interface for browsing jobs
Semantic search - Tell us what you're looking for in natural language ("remote Python jobs at startups") and our AI finds relevant matches
Smart tracking - We remember what you've seen so you never get the same job twice
One-tap apply - When you find a job you love, our AI can auto-fill applications for you. No hassle

How we built it

Architecture:

3 Python microservices (FastAPI) handling authentication, job scraping, and video delivery
Expo React Native mobile app with HLS video streaming
MongoDB Atlas with vector search for semantic job matching
Google Gemini for generating embeddings and powering our AI features
DigitalOcean Spaces CDN for low-latency video delivery worldwide

Video Pipeline: We built a custom pipeline that takes job descriptions and generates engaging videos:

LLM generates a script with character dialogue
Fish.audio converts text to speech with emotion
FFmpeg composites everything with karaoke-style captions
Videos are transcoded to HLS and served via CDN

Job Automation: Our headless service uses Playwright to scrape Greenhouse job boards and can even auto-fill applications using AI-generated responses tailored to each job.

Challenges we ran into

HLS streaming on mobile - Getting expo-video to play HLS streams reliably took significant debugging. We had to carefully manage player lifecycle, handle edge cases around visibility changes, and optimize prefetching to make scrolling feel smooth.

Vector search tuning - MongoDB's vector search is powerful but getting the right balance of numCandidates vs limit while filtering seen jobs required iteration. Too few candidates and relevant jobs get missed; too many and latency suffers.

Form automation - Greenhouse forms are surprisingly complex with React-Select dropdowns, conditional fields, and various input types. Our Playwright automation had to handle all these edge cases gracefully.

Accomplishments that we're proud of

Sub-second video starts - HLS + CDN edge caching means videos begin playing almost instantly
Seamless infinite scroll - Smart prefetching and player management makes the feed feel native
Full-stack AI integration - Gemini powers search, content generation, and application assistance
Production-ready architecture - Docker Compose deployment, health checks, proper error handling

What we learned

HLS is the way - For mobile video, HLS with CDN beats serving raw MP4s every time
Vector search is magic - Semantic search with embeddings creates a much better UX than keyword matching
Caching is everything - In AI pipelines, cache aggressively at every step
Mobile video is hard - Player lifecycle management, audio sessions, and performance optimization are non-trivial

What's next for ReelJobs

Personalized recommendations - Learn from swipe patterns to surface better matches
Company culture videos - Partner with companies to include authentic team content
Extending possibilities - Bring the ReelJobs experience to LinkedIn and other job boards