SkillForge AI

Inspiration:

I built SkillForge because I watched too many students around me choose careers they didn't actually want, usually because they didn't know what else was out there. A friend picked Computer Science because someone told him it "pays well," even though he hated coding. Another wanted to work in human rights but had no idea that was even a career path you could plan for.

Career guidance at most schools is either nonexistent or a one-size-fits-all list of "top jobs." I wanted something that actually asks who you are, what you're into, what you're good at, what matters to you, and builds a real path from there.

What It Does:

SkillForge is a career pathfinder that helps students figure out what they actually want to do, then shows them how to get there.

-Adaptive quiz, starts with your grade level and interests, then asks follow-up questions shaped by what you already said. No two quizzes are the same. -Real career matches, not a generic top-three list. The AI looks at your actual answers and suggests careers that fit your personality, values, and strengths.

Grade-aware roadmaps, whether you're in 9th grade or already working, the plan starts from where you are. Each phase has specific skills, projects, free resources, and milestones.
Works offline too, if the AI is down, a local engine with 30+ careers still gives you solid recommendations and full roadmaps.

How I Built It:

Frontend: Plain HTML, CSS, and vanilla JavaScript. I wanted it fast, lightweight, and something any student could open in a browser without installing anything.

Design: I went with a route-map aesthetic, warm paper tones, forest-green accents, dashed lines showing your path from "where you are" to "where you're going." It feels like a field guide, not a corporate dashboard.

AI: Groq API with Llama 3.3 70B. I originally tried Gemini but Google blocked the key within minutes (their new AQ. format is broken). Groq is faster and cheaper anyway.

The tricky part was the prompts. Early versions kept defaulting to "Software Developer" and "Data Scientist" for everyone, even someone who said they wanted to fight for human rights. I had to explicitly tell the AI: do not recommend tech unless the student actually mentioned coding. I also made it generate realistic salaries (entry-level nonprofit work is $35k–$55k, not $100k) and career-specific roadmaps (no "learn Git" for aspiring social workers).

Local fallback engine: I built a complete scoring system with interest tags, category detection, and template roadmaps for 7 career areas. When the AI fails, the app doesn't break, it just uses local data.

What I Learned:

Prompt engineering is most of the work. Getting the AI to stop being lazy and actually match careers to people took dozens of iterations.
Fallbacks are non-negotiable. APIs fail. Keys get blocked. If your app depends entirely on an external service, it's not a real product.
Students need honest information. Inflated salaries and generic advice do more harm than good. Being upfront about low pay in advocacy work builds trust.

Challenges I faced:

The infinite fallback loop: When the AI failed, my duplicate-question logic created an endless loop of the same fallback questions. I fixed it by tracking original IDs separately from timestamped ones.
API key roulette: Google's key format change broke Gemini integration 15 minutes after I created the key. Pivoted to Groq mid-build.
Roadmap relevance: Early AI outputs for non-tech careers still included "learn Git repositories." I had to explicitly ban generic tech skills and force career-specific resources in the prompts.

What's Next: