PixelForge — The Resume That Writes Itself

ai usage
otp
github repo
signin/signup dashboard
resume
2fa
profile/security

Inspiration

I'm a second-year CSE student at KL University. I have 3 deployed projects, 150+ open source PRs, and a GitHub that updates almost every day. But when I applied for internships, I was sending out the same 1-page resume as everyone else — generic bullet points, no real metrics, no proof of work.

The problem isn't that students don't have skills. It's that their resumes don't show them. A student who built 10 projects and solved 500 LeetCode problems looks identical on paper to someone who did nothing. First-generation students with no network, no career counseling, and no alumni mentorship are hit hardest.

I wanted to fix this.

What it does PixelForge is a Live Developer Identity Platform. Instead of manually writing a resume, you connect your GitHub, LeetCode, and coursework. PixelForge:

Syncs your real work — repos, commits, languages, problem-solving stats
Extracts skills and achievements — what you actually built, not what you claim
Generates a tailored resume — paste a job description, get a 1-page resume in 30 seconds
Writes real bullets — "Reduced API latency by 40%" instead of "hardworking team player"
AI twin for recruiters — recruiters can chat with an AI version of your profile for quick screening

How I built it

Frontend: Next.js 15 (App Router), Tailwind CSS, Clerk for auth
Backend: FastAPI, PostgreSQL (Neon), SQLAlchemy
AI: Gemini 3.1 Pro + Groq (multi-model router with fallback)
Embeddings: sentence-transformers for semantic matching
Deploy: Render (backend) + Netlify (frontend)

The AI layer uses a router that picks the best model per task — Gemini for resume rewriting, Groq for cost-efficient fallback. Every AI call is logged to an AI usage dashboard so the user can see cost, latency, and token usage.

Challenges I faced

GitHub API rate limits — unauthenticated calls are capped at 60/hour. Added a token for 5,000/hour.
Semantic matching — matching repos to job descriptions needed embeddings, not keyword search.
Multi-model routing — building a fallback chain so the product doesn't break if one API goes down.
AI usage transparency — every hackathon requires disclosing AI usage. I built a dashboard that tracks every call.

What I learned

Building with real tools beats building in a sandbox.
AI is a force multiplier. The AI wrote the boilerplate. I wrote the logic and made the product decisions.
Shipping matters more than perfecting.

What's next