Rectify Learn
Co-founder & COO · 2025–26- 150+active users
- 4startup programs: Microsoft, GitHub, Notion, Miro
- SilverAI category, among 200+ teams
An AI study platform that turns a student's own notes and PDFs into quizzes, flashcards and explanations, then schedules each review around what they are about to forget. I ran product, operations and partnerships, and led frontend–backend API integration.
- UploadPDF, DOCX, images
- ParseDocling to Markdown; Tesseract OCR
- Chunk & embedRecursive chunks + metadata; all‑MiniLM‑L6‑v2
- IndexPostgreSQL + pgvector, HNSW
- GenerateRAG over top-5 chunks: quizzes, flashcards, explanations
- ScheduleMastery Score 0–100 + modified SM‑2
From our technical whitepaper, which won Silver (AI category) at Blockchain Olympiad Bangladesh 2025 and a nomination to the International Data Science Olympiad.
3 design decisions
- Grounded in the student's own material. Every question is generated from retrieved passages of what they uploaded, so it stays on their syllabus.
- Reviews timed by mastery, not a fixed calendar. A 0–100 Mastery Score blends accuracy, response time and self-rated confidence; a modified SM-2 sets the next interval as I(n) = I(n−1) × EF.
- Designed for teenage learners. For SSC and HSC students: non-competitive progress, non-stigmatizing feedback, data minimization, and a review schedule learners can see and override.
- React
- Django
- PostgreSQL
- pgvector
- HNSW
- Docling
- Tesseract
- SentenceTransformers