About The Company

CollegeDoors, founded by IIT and IIM graduates, is dedicated to improving education for engineering and medical aspirants in India. The platform offers high-quality questions, intuitive tools, and personalized service to support teachers and students. By reducing teachers' operational workload, CollegeDoors enables them to focus on core teaching. Continuously evolving based on feedback, the company aims to foster educational excellence nationwide, empowering students and teachers in every town to build their success stories.

‍

Problem Statement

The customer had to use low resolution images in their test series which was giving it an unprofessional look and feel causing customer churn to competitors.

‍

Key Challenges

Mathpix API was expensive (for the budget of the project), so we merged multiple images together (with placeholder text) to reduce 3P API cost.

‍

Approach

Step 1: Research - We benchmarked OCR models available for (i) free (Tesseract, Pix2Text) and (ii) paid (Mathpix and Amazon Textract). This use case consisted of extracting mathematical equation and chemical formula’s, Mathpix was the most accurate model with >98% accuracy.

Step 2: Formatting - The client requested the equations to be output in MathType and MathML format that exactly matches their product UI so we converted LaTeX into MathML and MathType

Step 3: Development - We developed a python program that fetches the images from database, merges multiple images together, calls Mathpix API to extract text, equations and figures separately. The LaTeX is converted to MathML and MathType formats and updated in-place in the database and the corresponding image assets uploaded to S3 bucket

‍