Padhik
Excel in both your exams and Career.
The problem Padhik solves
The Problem It Solves:
Keralan students face a dual dilemma in their academic journey:
- Academic Struggles (Low CGPA, Ineffective Exam Prep)
University curricula are outdated and not aligned with current exam formats.
Students lack personalized support for doubt-solving and revision.
YouTube and other online resources are unstructured and distracting, making self-study chaotic.
Result: Low CGPA, poor fundamentals, and exam anxiety.
- Career Confusion (Skill Gaps, No Industry Readiness)
Despite having degrees, many students are not job-ready.
They are unaware of local job/internship opportunities (like those in Technopark, Infopark).
Students struggle to identify missing skills or where to learn them.
Existing career guidance is generic and lacks personalization.
What Project Padhik Enables:
- Boost Your Grades (XMAI Wing)
Upload lecture notes or PYQs → get AI-powered explanations
No more last-minute panic or scattered revision
Master concepts the smart way with curated video learning
2.Build Your Career (Career Navigator)
Upload resume + GitHub → get skill gap analysis + career roadmap
Discover localized opportunities based on your current strengths
Receive custom learning paths (from Coursera, KKEM, local workshops, and YouTube)
Challenges we ran into
Challenges I Ran Into
- Integrating AI Models with Frontend (Flask ↔ Svelte)
Our biggest challenge was seamlessly connecting the AI backend (Flask) with the modern frontend (Svelte). We faced:
CORS issues while sending fetch requests.
File upload handling (PDFs, resumes) from Svelte to Flask was initially inconsistent.
Solution:
We carefully configured CORS headers and used FormData() for handling uploads. We also debugged using curl and browser network tools to pinpoint exact request/response mismatches.
- Resume Parsing & Skill Extraction
Parsing resumes and extracting meaningful career data using Python was complex. Many resumes had inconsistent formats, making it hard for AI to identify relevant skills or GitHub links.
Solution:
We tested multiple libraries (PyMuPDF, pdfminer, spacy) and finally built a lightweight rule-based parser combined with GPT to extract the most useful sections.
- Maintaining AI Context for PDF Questioning
For the study assistant (XMAI), we had to chunk large PDFs and retain question-answer accuracy across multiple pages without losing context.
Solution:
We implemented langchain for vectorized search and faiss indexing, allowing semantic search across chunks. This gave us much more accurate, context-aware answers.
Tracks Applied (1)