the AI-powered search engine for ur digital hoarding habits.
ok real talk: ur "Saved" folder on instagram is literally a graveyard. u got a 3-year-old sourdough recipe, a workout u SWORE u'd do, and a "life hack" for a toaster u don't even own.
Reel-Mind uses AI to actually categorize these reels so u can find them again. think of it as a personal librarian for ur ADHD-fueled 2am scrolling sessions. except now it has a login page so ur saved reels are actually YOURS and not just vibes floating in a database.
the live demo is down. instagram killed the yt-dlp approach that powered the whole thing and i'm not in a position to fight meta right now. the backend, the auth, the AI pipeline — all of it still works. just the video download layer got bodied.
but here's the thing: the codebase is actually pretty clean. the architecture is solid. so instead of letting it rot in a private repo, i'm open-sourcing it.
if u want to self-host ur own instance, fix the download layer, or build on top of what's here — go for it. everything u need is in
SELF_HOSTING.md.
still a freshman. still figuring things out. the codebase is maybe slightly less spaghetti than before but like... no promises.
- The Code: held together with duct tape, stackoverflow, and Claude (the AI not the person. well. both actually)
- The Logic: it works on my machine and that's good enough for now
- The UI: ok tbh this one actually looks kinda clean now ngl
there's a proper sign in / create account flow. ur reels are actually tied to ur account instead of just... existing in the void. built with supabase auth.
u can view ALL ur categorized reels in one place. each card shows:
- the summary of what the reel is actually about (so u don't have to watch it again to remember)
- the categories gemini slapped on it (Technology, Software, etc)
- the tags for more specific vibes (#Dangerzone #sandbox #malware prevention etc)
- the date u saved it
- a button to go view it on instagram
- a delete button for when u realize u don't actually care about that reel anymore
there's a lil settings gear icon. it exists. that's all i'll say.
was live. on the internet. for real people. good while it lasted.
- The Handover: u drop a url. app says "bet"
- The Heist (
yt-dlp): backend sneaks in and downloads the reel. digital repo man for ur memes - The Interrogation (Gemini AI): i basically tell gemini "bro just tell me what this is and put it in a category, i am NOT rewatching 47 saved reels"
- The Membership Check: supabase auth makes sure ur actually logged in before saving anything
- The Archive (Supabase): categories + tags + summary get saved to ur personal library in the db
- The Cleanup: backend deletes the downloaded video immediately after. no hoarding on the server
- The Delivery: fastapi sprints back to the frontend and ur reel shows up all neat and categorized
- Python & FastAPI — the engine room. still fast, still forgiving when i forget a colon
- yt-dlp — heavy lifter that grabs the videos. legend (RIP for now)
- Gemini AI — the actual brain. im just the guy who asked nicely
- Supabase — handles the db AND auth. doing double duty. RLS enabled so ur data is actually urs
- HTML/CSS/JS — frontend looking cleaner than ever (claude "assisted" 🙏)
- Railway — where the backend lived. real deployment hours
- GitHub Pages — served the frontend. fast, free, zero config
- Hopes and Dreams — down to like 40% of the codebase now. progress.
the live demo is down but u can run ur own instance. the whole setup is documented in SELF_HOSTING.md — supabase config, railway deploy, frontend setup, the works.
quick start:
# 1. clone the chaos
git clone https://github.com/msjabata25/Reel-Mind.git
# 2. get the python stuff
pip install -r requirements.txt
# go make a snack, it takes a sec
# 3. set up ur .env
# SUPABASE_URL=your_url
# SUPABASE_KEY=your_service_role_key
# ALLOWED_ORIGIN=your_frontend_url
# 4. set up ur frontend config
cp config.example.js config.js
# fill in ur supabase + backend URLs
# 5. fire it up
uvicorn main:app --reload
# 6. go to localhost:8000 and prayfull guide with supabase table setup, RLS policies, instagram cookies, and troubleshooting → SELF_HOSTING.md
- yt-dlp fragility: instagram actively fights scrapers. this is ultimately what shelved the project. if u fix this layer, lmk
- gemini json parsing: gemini occasionally wraps its response in markdown fences despite being explicitly told not to. working on stripping those before parsing
- session expiry: supabase tokens expire after an hour. if u get a weird auth error after leaving the tab open, just sign in again
- mobile layout: it's... fine? mostly? squint a little
- auth (sign in / create account)
- saved reels library with categories + tags
- delete saved reels
- summary display on reel cards
- search / filter ur saved reels by tag or category
- RLS policies — ur data is actually urs
- deployed (Railway backend + GitHub Pages frontend)
- automatic video cleanup after processing
- open-sourced for self-hosting
- a download layer that actually survives instagram's bot detection
- TikTok + YouTube Shorts support (url validation already done, just needs testing)
- proper logging instead of
print("wtf happened") - react native app — mobile version
- maybe a chrome extension?? idk that sounds hard
- survive uni while also not getting left behind by the job market
- internship (still doubtful but the dream lives on)
if ur a senior dev: please look away. i am begging u
if ur a fellow student: we suffer together. PRs welcome, judgment not
if u cracked the instagram download problem: u are legally required to open a PR
built with ❤️, ☕, way too many gemini api calls, and a very confused look on my face by msjabata25