Client: TestnTrack
Live Site: https://www.testntrack.com
Objective: Automate grading of OMR and handwritten answers, cut costs, improve turnaround.
Tech Stack: PyTorch, Detectron2, React Native, AWS Fargate, Vision Transformers, LLM, ELK Stack
Overview:
Manual teacher-led grading caused a 3-day delay in results with rising costs. Scaling was not feasible.
Challenge:
Manual, slow, expensive workflows that hindered performance at scale.
Solution:
- OMR Engine: Vision model to classify and digitize bubble sheets.
- Subjective Engine: Transformer + LLM for handwriting recognition and scoring.
- Mobile data ingestion and offline queuing.
- Deployed via AWS microservices with API dashboard access.
Key Results:
- Result turnaround: 72h → 4h
- Cost per sheet: ₹20 → ₹3
- Accuracy increased to 98.5%
- Monthly cost savings: ₹10L+
