EduPath AI — From Curiosity to Campus, Powered by AI

💡 Inspiration

Every year, over 1.3 million Indian students plan to study abroad. Most navigate this journey completely alone — juggling Google searches, confusing visa guides, Excel sheets for deadlines, and a last-minute scramble to fund it all. We watched classmates make uninformed decisions about universities and loans simply because no single platform guided them through the full journey. When we read the Poonawalla Fincorp problem statement, it resonated immediately. The gap wasn't just a product gap — it was a trust gap. Students don't apply for education loans early because no platform earns their trust first. I decided to build that platform.

🔨 How I Built It

I built EduPath AI from scratch using Python and Streamlit for the frontend, and the Groq API (LLaMA 3.3 70B) as our AI backbone. The platform follows a three-layer architecture:

Layer 1 — AI Engagement (Top of Funnel)

Twelve AI-powered tools — Career Navigator, Admission Predictor, Timeline Generator, SOP Generator, and Scholarship Finder — all take the student's persistent profile as context and generate structured, personalized outputs via carefully engineered prompts.

Layer 2 — Financial Tools (Middle of Funnel)

The ROI Calculator projects net career gain using:

$$\text{ROI} = \frac{(\text{Post-Course Salary} - \text{Pre-Course Salary}) \times \text{Years} - \text{Course Cost}}{\text{Course Cost}} \times 100$$

The Loan Eligibility Engine uses NBFC-standard FOIR logic:

$$\text{Max EMI} = (\text{Monthly Income} \times 0.50) - \text{Existing EMI}$$

$$\text{Max Loan} = \text{Max EMI} \times \frac{(1+r)^n - 1}{r \cdot (1+r)^n}$$

Where $r$ is the monthly interest rate and $n$ is the repayment period in months. Interest rates are tiered by credit score: 9.5% (750+), 11% (700–749), 13% (650–699), 15.5% (<650).

Layer 3 — Conversion (Bottom of Funnel)

A 4-step AI-assisted loan application form, pre-filled from the student's profile, generates a personalized loan offer letter via LLaMA 3.3. A 30-day automated email nudge sequence simulates a zero-human-intervention growth loop— the bonus challenge from the problem statement — covering 5 journey stages: Onboarding → Activation → Engagement → Nurture → Conversion.

I also built a complete growth engine with XP gamification (6 badge levels, 0–400+ XP), daily streaks with milestone bonuses at days 3, 7, and every 10 days, unique referral codes, and context-aware smart notifications triggered by profile completeness, deadline proximity, and loan eligibility results.

📚 What I Learned

• Prompt engineering is a discipline. Getting LLaMA 3.3 to output clean, consistently structured markdown across 12 different tools required explicit section headers, tone instructions, and format constraints in every system prompt.

• Rule-based logic complements LLMs Financial calculations like

Built With

• cloud
• cloud:
• control:
• database:
• datetime
• deployment
• engine)
• file-based
• git
• groq
• hashlib
• hosting)
• inference
• json
• languages:-python-3.12-frameworks:-streamlit-(web-app-&-ui)-apis:-groq-api-?-llama-3.3-70b-versatile-(llm-inference)-libraries:-groq
• llm
• local
• os
• persistence)
• plotly
• profile
• python-dotenv
• storage
• streamlit
• user
• version