How I Built an AI Co-Founder Matching Platform Assisted By GitHub Copilot

This is a submission for the GitHub Finish-Up-A-Thon Challenge

What I Built

FoundrGeeks is a global platform built around one core belief:

The people you build with matter more than anything else.

FoundrGeeks core values:

Networking: you discover builders around you who share the
same interests, the same problems, and the same drive. Not random
strangers, but people your profile is actually compatible with,
surfaced through AI matching.

Building your team: whether you are looking for a co-founder,
a skilled developer, a designer, or a product person, the platform
helps you find the right fit based on real compatibility, not
just keywords on a profile.

Hiring: early stage founders and skilled professionals can
find each other in the same space. A developer looking for a
dream project to join and a founder looking for that exact
developer are both here, and the platform connects them.

Core Features:

• 🤖 AI-Powered Talent Matching: Uses NVIDIA Text Embeddings and LangChain to perform cosine similarity vector searches that mathematically calculate co-founder compatibility based on bios, skills, and goals
• 🚀 Project Showcase Board: Founders can pitch their startups and recruit the exact roles they need
• 💬 Real-Time Discussion Forums A chatty, Facebook-style community space for collaboration, feedback, and networking
• 🔔 Smart Notifications: Real-time alerts for matches, replies, and new connections

Tech Stack: Next.js, Node.js, Express, MongoDB, NVIDIA Text
Embeddings, LangChain, JWT Auth

🔗 GitHub: https://github.com/Isreal-Oparanti/Launchpad

🔗 Website: https://foundrgeeks.com

Demo

The Comeback Story

This project has a real history and it shows in my commit log.
It started a last year August. Early commits like "fix cors issues",
"fix auth page", and "added talents" go back months.
I was building it in phases, pushing when I could, then going quiet
for weeks. A typical side project story.
At some point I hit a wall. The matching system existed but it
was using OpenAI embeddings and it was slow and not accurate
enough. The repo sat there with a commit that
said "Preparing for launch" that never actually launched.
What was missing:

• The AI matching was unreliable and slow (OpenAI was the wrong tool for this)
• UI was inconsistent across pages, lots of small bugs
• No real demo data to show anyone
• The whole thing just felt unfinished When I saw the GitHub Finish-Up-A-Thon challenge, it felt like it was written for exactly this project.

What changed:
First thing I did was rebuild the matching engine from scratch.
I switched from OpenAI to NVIDIA Text Embeddings with LangChain
and a cosine similarity vector search. The difference was
immediately obvious. Faster, more accurate, and the match scores
actually made sense.

Then I Improved the discussion forums. Threaded replies, real-time
comment counts, pagination, delete controls, the full thing.
Then I went through the UI properly. Profile pages, talent cards,
project cards with real logos and cover images, navigation fixes,
avatar rendering bugs. The kind of polish work you keep pushing
back when you are chasing bigger features.

Finally I built proper demo seed scripts so the platform actually
looks and feels alive when someone opens it for the first time.

The commit that marks the finish line:

"Finish project features: AI Matching, Discussion Forums, Gen-Z avatars"---> with GitHub Copilot as co-author

My Experience with GitHub Copilot

Honestly the biggest thing Copilot did was keep me from stopping.
When I was working alone on a side project that has already
been "almost done" for months, the hardest part is yet perfected and lots of small-small tasks to do which github copilot took care of.

The NVIDIA matching engine was the most technically heavy
part of this whole build. Getting the LangChain pipeline right,
structuring the vector search, handling edge cases when user
bios were too short to embed properly. Copilot suggested
patterns I wouldn't have found quickly on my own and kept me
moving through the hard parts.

The discussion forum backend was one of those things I kept
avoiding because wiring up full CRUD with proper permissions,
reply counts that stay accurate, and delete controls that only
show for your own replies felt like a lot. Copilot helped me
get through it in one session instead of spreading it across
multiple days.

Debugging was where it really earned its keep. A few UI bugs
like logo images not loading and avatar rendering glitching on
certain pages had me stuck. Copilot helped me trace the root
cause quickly both times.

The honest take: I would have finished this eventually
without Copilot. But not by the deadline.

FoundrGeeks is finally live.
Shout out to Githup Copilot.