GitTogether

Dating wrapper on top of GitHub

my yc caliber idea

hidden formula: 100 users, $100/month, $10k MRR

Normal dating apps:

• Fake bios
• Vibes > substance
• Zero signal

GitHub = high-signal identity

• Skills
• Consistency
• Obsession
• Taste (tech stack, OSS, side projects)

User Flow

1. Login with GitHub OAuth

2. App auto-builds:

   • Skill graph
   • Interest vectors
   • Personality proxies (from behavior)

3. You swipe / match / connect

4. Chat unlocks only after mutual intent

What Data You Pull From GitHub (IMPORTANT)

Raw Inputs

• Repos (own + starred)
• Languages used (%)
• Commit frequency (time-of-day too)
• OSS contributions
• README tone (serious vs meme)
• Issues / PR comments (optional, risky)
• Topics/tags

Derived Signals (this is your secret sauce)

• Builder vs Talker

  • Commits > stars

• Depth vs Breadth

  • Few langs deeply vs many shallow

• Solo vs OSS

• Night owl vs morning grinder

• Experimental vs production mindset

Matching Logic (don’t do dumb cosine similarity only)

Layered Matching

1. Hard Filters

   • Location / remote
   • Age range (legal stuff)
   • Intent (dating / collab / both)

2. Skill Affinity

   • Language overlap (weighted)
   • Domain overlap (web, infra, crypto, ML)

3. Mindset Compatibility

   • OSS heavy ↔ OSS heavy
   • Indie hacker ↔ indie hacker
   • ICPC grind ↔ ICPC grind

4. Contrast Bonus (important)

   • Frontend ↔ backend
   • Builder ↔ designer
   • Hardcore ↔ soft-skilled

Perfect similarity is boring.

Chat & Interaction Rules (this decides success)

Rules to avoid degeneracy:

• No instant DMs

• Unlock chat only after:

  • Mutual like
  • OR shared repo discussion

• First message prompt is auto-generated

  “You both starred X and use Rust. What do you like/hate about it?”

Services

• GitHub API (batched + cached)
• Background jobs (profile re-sync)
• Vector DB (optional later)

Data Model (simplified)

User
- github_id
- skill_vector
- activity_score
- intent

Match
- user_a
- user_b
- score
- created_at

Message
- match_id
- sender
- content

Privacy & Safety (skip this = dead product)

Hard rules:

• No private repos
• No commit diffs shown
• Opt-in visibility
• “Hide activity timing” toggle

Also:

• Women will be minority initially → protect them
• Rate limit likes
• Soft shadow bans for creeps

---

Cold Start Problem (this will kill you if ignored)

Hacky but effective:

• Invite-only alpha
• Target:
  • OSS contributors
  • Hackathon folks
  • Twitter/X devs

Let density build in one niche first:

“Rust + OSS people in India / EU”

Don’t go global day 1.

---

Monetization (don’t lie to yourself)

NOT subscriptions initially.

Possible:

• Boost visibility (soft)
• Events / meetups
• Hiring funnel later (careful)
• Premium insights (“who viewed your profile”)

If you monetize too early → trust dies.

pull the github profiles, and other coding profiles once pulled, get people from same city together then u need to apply machine learning algorithms which work on data similarity

other things like movie taste,

well things like pull tech stack like if one is writing solana dev, ts dev

someone is doing projects or dsa

cp people, keep similar ratings close to each other

pull data once in 24 hours for each profile like once you sign up, get the time at that moment then every 24 hours for that profile only, need to pull data

data pulled (repos, activity, location)

tech stacks overlap, contribution cadence, timezone/city, language preference, collaboration patterns.

GitHub login, city filter, simple similarity scoring, weekly match email, feedback loop.

Users

reach out to people

market as much as you can

do personal on boardings

More of search problem than persuasion problem

Charge real money early, feedback over revenue

Use targeted personal outreach

Launch early

Study early users closely

Experiment and fail fast, talk to users

Early users not only give feedback but they also shape how your growth takes place

First version shouldn't be min viable product; But it should be min evolvable product